Cow perinatal period energy metabolism regulation and control method and system

Abstract

The invention provides a dairy cow perinatal period energy metabolism regulation and control method and system, and relates to the technical field of livestock breeding, and the method comprises the steps: obtaining perinatal basic information, and carrying out the structural processing, and obtaining perinatal data information; extracting association features among entities in the perinatal data information through a graph neural network, and generating a multi-modal knowledge graph; performing characterization learning on associated features in the multi-modal knowledge graph by using a deep belief network in combination with historical metabolic data and field expert experience, and generating a grain supply rule base; and based on the grain supply rule base, generating a daily ration formula through a reinforcement learning algorithm.

Description

Technical Field

[0001] The present invention belongs to the technical field of livestock breeding, and particularly relates to a method and system for regulating energy metabolism during the periparturient period of dairy cows. Background Art

[0002] The periparturient period is a critical stage in the production cycle of dairy cows. Energy metabolism disorders are likely to lead to diseases such as ketosis and fatty liver. The existing regulation methods mainly rely on limited physiological indicators and empirical feed formula adjustments. Although they can meet the requirements to a certain extent, there are obvious deficiencies: In terms of data processing, the existing methods only focus on limited indicators such as blood BHB and non-esterified fatty acids, ignoring the fusion of multi-source data such as environmental parameters and continuous physiological signals, and cannot comprehensively reflect the energy metabolism state of dairy cows. In terms of feature extraction, the existing technology is relatively single and it is difficult to capture the deep features and complex relationships in the data. In terms of feed formula formulation, the existing methods are mostly based on experience. Although there are nutritional standards for reference, they do not combine the individual differences of dairy cows and environmental changes, resulting in insufficient accuracy. It is difficult to dynamically adjust the feed formula according to the real-time metabolism state of dairy cows and cannot meet their dynamic nutritional needs in a timely manner. Summary of the Invention

[0003] Based on this, it is necessary to provide a method and system for regulating energy metabolism during the periparturient period of dairy cows that can solve the above technical problems.

[0004] In the first aspect, the present application provides a method for regulating energy metabolism during the periparturient period of dairy cows, including:

[0005] Obtaining periparturient basic information and performing structured processing to obtain periparturient data information;

[0006] Extracting the association features between entities in the periparturient data information through a graph neural network to generate a multimodal knowledge graph;

[0007] Combining historical metabolic data and domain expert experience, using a deep belief network to perform representation learning on the association features in the multimodal knowledge graph to generate a feeding rule library;

[0008] Based on the feeding rule library, generating a daily ration formula through a reinforcement learning algorithm; the daily ration formula is used to regulate energy metabolism.

[0009] In one of the embodiments, this method further includes:

[0010] Extracting the multimodal time-frequency domain features of periparturient basic information through time series coding and high-dimensional spectral analysis technology, and constructing a three-dimensional spatio-temporal feature matrix based on the multimodal time-frequency domain features; the periparturient basic information includes environmental parameters, continuous physiological signals, and metabolite indicators;

[0011] Based on the three-dimensional spatio-temporal feature matrix, a metabolic state prediction model is constructed using transfer learning technology;

[0012] The metabolic state prediction model is used to generate adjustment information, and the feeding rule base is updated based on the adjustment information.

[0013] In one embodiment, a three-dimensional spatio-temporal feature matrix is constructed based on multi-modal time-series frequency-domain features, including:

[0014] The three-dimensional spatio-temporal feature matrix is constructed by the following formula:

[0015]

[0016] where τ represents the feature fusion function, LSTM represents the time-series feature, FFT represents the frequency-domain feature, P(t) represents the physiological signal time-series data, M(t) represents the metabolite frequency-domain feature, h p / h m / h c represents the feature dimensions of each modality.

[0017] In one embodiment, using transfer learning technology to construct a metabolic state prediction model, including:

[0018] Initializing the model parameters using the cow parturition dynamics simulation data set;

[0019] Using a three-dimensional convolutional neural network to fuse time-series - frequency-domain features, and using the following formula to output the metabolic risk probability and fine-tune the model:

[0020] P(y=1|T(t))=σ(W f ·F final (t)+W c ·C clinical )

[0021] where F final represents the fused feature vector, C clinical represents the clinical detection index, and σ represents the Sigmoid activation function.

[0022] In one embodiment, the method further includes:

[0023] Calculating the dynamic energy metabolism index according to the obtained perinatal basic information;

[0024] Updating the feeding rule base according to the dynamic energy metabolism index, where the dynamic energy metabolism index is dynamically calculated by the following formula:

[0025]

[0026] where C BHB represents the plasma β-hydroxybutyric acid concentration, CTrp represents the plasma tryptophan concentration, CFU Cellulolytic represents the colony forming unit of cellulose-degrading bacteria, CaO represents the dietary calcium oxide content, CaD represents the total dietary calcium content, NEFA represents the non-esterified fatty acid concentration, and MUN represents the urinary urea nitrogen concentration.

[0027] In one embodiment, the method further includes:

[0028] using the following formula to optimize the weight parameters of the dynamic energy metabolism index formula:

[0029]

[0030] where J(θ) represents the weight parameter, feed represents the feed formulation parameter, y represents the production performance and health indicators, λ represents the rule stability penalty coefficient, and ||θ new -θ old || 2 represents the Euclidean distance quantifying the difference between the old and new parameters.

[0031] In one embodiment, the method further includes:

[0032] adjusting the dietary formula according to the following steps:

[0033] According to the BHB and parity information of prepartum cows, adjust the dietary energy concentration in the dietary formula in stages, and add dry matter with a vitamin D3 content of 2000 - 3000 IU / kg and 0.3% - 0.5% magnesium;

[0034] According to the results of rumen fluid metagenomic analysis, adjust the ratio of dietary fiber to starch in the dietary formula;

[0035] Dynamically set the BHB threshold according to parity, and add bypass fat to the dietary formula corresponding to cows with BHB above the threshold within 7 days after calving.

[0036] In one embodiment, the method further includes:

[0037] Adjust the dietary formula of postpartum cows according to the following stages:

[0038] In the first week after calving, adjust the dietary energy supply concentration to: 1.45 - 1.50 Mcal / kg DM;

[0039] In the second week after calving, adjust the dietary energy supply concentration to: 1.51 - 1.55 Mcal / kg DM;

[0040] From the third week after calving to the end of the lactation peak, adjust the dietary energy supply concentration to: 1.56 - 1.60 Mcal / kg DM.

[0041] In one embodiment, the method further includes:

[0042] Adjust the dietary formula for dairy cows in the peak lactation period according to the following rules:

[0043] Maintain the dietary calcium concentration in the range of 0.90%-1.10% on a dry matter basis;

[0044] When the dietary energy supply concentration ≥ 1.55 Mcal / kg DM, add 250-300 g / d of bypass fat;

[0045] When it is detected that plasma NEFA > 1.2 mmol / L, add 500 IU / d of vitamin E;

[0046] When it is detected that MUN > 15 mg / dL, reduce the dietary crude protein to 16%-17%.

[0047] In a second aspect, the present application also provides a periparturient energy metabolism regulation system for dairy cows, including a periparturient information processing module, a periparturient atlas generation module, a feeding rule library generation module, and a dietary formula generation module:

[0048] The periparturient information processing module is used to obtain periparturient basic information and perform structured processing to obtain periparturient data information;

[0049] The periparturient atlas generation module is used to extract the association features between entities in the periparturient data information through a graph neural network to generate a multimodal knowledge graph;

[0050] The feeding rule library generation module is used to combine historical metabolic data and domain expert experience, and use a deep belief network to perform representation learning on the association features in the multimodal knowledge graph to generate a feeding rule library;

[0051] The dietary formula generation module is used to generate a dietary formula based on the feeding rule library through a reinforcement learning algorithm.

[0052] The above-mentioned method and system for regulating energy metabolism during the perinatal period of dairy cows obtain and structurally process the basic perinatal information, generate a multi-modal knowledge graph using a graph neural network, and fuse multi-source data such as environmental parameters, continuous physiological signals, and metabolite indicators to solve the problem that the existing technology only focuses on limited indicators in data processing and cannot comprehensively reflect the energy metabolism state of dairy cows; extract associated features through a graph neural network, perform representation learning using a deep belief network, and deeply mine the deep features and complex relationships in the data to overcome the deficiency of the existing technology in relatively single feature extraction; based on the generated feeding rule library, dynamically generate a diet formula through a reinforcement learning algorithm, and can accurately adjust the feed according to the real-time metabolism state and environmental changes of dairy cows to solve the problem that the existing technology formulates feed formulas only based on experience and nutritional standards and does not combine individual differences and environmental changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0054] Figure 1 It is a flowchart of a method for regulating energy metabolism during the perinatal period of dairy cows according to the present invention;

[0055] Figure 2 It is a schematic structural diagram of a system for regulating energy metabolism during the perinatal period of dairy cows according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] In order to make the objectives, technical solutions, and advantages of the present application clearer, the following further elaborates on the present application in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0057] The method and system for regulating energy metabolism during the perinatal period of dairy cows of the present application have an implementation environment including a dairy farm, a laboratory, and a data processing center, and involve data acquisition devices, data processing devices, data storage devices, and network devices. Through the collaborative work of these hardware, the monitoring, analysis, and regulation of the energy metabolism of dairy cows during the perinatal period are realized. When it is necessary to regulate the energy metabolism of perinatal dairy cows, the data acquisition device starts to work, monitors the environmental parameters, physiological signals, and metabolite indicators of dairy cows in real time, and transmits these data to the data processing device through the network device for processing and analysis. The data processing device generates a multi-modal knowledge graph and a feeding rule library, and then generates a diet formula through a reinforcement learning algorithm, and transmits the formula information to the execution device to achieve precise feeding.

[0058] In one embodiment, as Figure 1 shown, a method for regulating energy metabolism during the periparturient period of dairy cows is provided. In this embodiment, an example is given where this method is applied to a terminal. It can be understood that this method can also be applied to a server and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0059] S101, Obtain periparturient basic information and perform structured processing to obtain periparturient data information.

[0060] Among them, the periparturient basic information includes multi-dimensional heterogeneous data during the periparturient period of dairy cows (3 weeks before calving to 3 weeks after calving), which can be environmental parameter information such as temperature, humidity, light intensity, air quality, continuous physiological signals such as body temperature dynamic monitoring sequences, heart rate variability HRV time series data, rumination activity rhythms, metabolite indicators such as biochemical test values of blood β-hydroxybutyric acid BHB, non-esterified fatty acid NEFA, and blood urea nitrogen MUN concentration. Convert the original heterogeneous data into a machine-parsable standardized data form, remove outliers through data cleaning, such as extreme values caused by sensor noise, fill in missing data, such as based on time series interpolation or machine learning prediction, and perform standardized conversion such as unifying dimensions and normalization processing to obtain periparturient data information, and convert fragmented data into a machine-parsable standardized format to support multi-modal feature fusion.

[0061] S102, Extract the association features between entities in the periparturient data information through a graph neural network to generate a multi-modal knowledge graph.

[0062] Among them, to mine the complex association relationships between entities in periparturient data, a graph neural network GNN can be used for modeling. The nodes include dairy cow individuals, physiological indicators, and environmental factors, and the edges are the association relationships between entities. The association features of each entity are obtained using the time series features and frequency domain features. At the same time, domain knowledge such as the impact of parity on energy requirements can be encoded as graph node attributes, and the non-linear associations between data can be explicitly expressed through the graph structure, providing a global perspective for metabolic state prediction.

[0063] S103, Combine historical metabolic data and domain expert experience, and use a deep belief network to perform representation learning on the association features in the multi-modal knowledge graph to generate a feeding rule library.

[0064] Among them, learning metabolic laws from the knowledge graph and forming interpretable feeding rules can be achieved by using a deep belief network for unsupervised pre-training, such as extracting high-order features of the graph through a restricted Boltzmann machine, and supervised fine-tuning, such as optimizing network parameters by combining historical metabolic data. The attention mechanism is used to identify key features, such as the need for intervention when β-HBA > 1.2 mmol / L on the 7th day after parturition. The identified features are transformed into rules understandable by domain experts. For example, if the parity ≥ 3 and MUN > 15 mg / dL, the dietary protein is reduced. The data-driven model is combined with expert experience to improve the credibility and applicability of the rules.

[0065] S104, based on the feeding rule library, generate a dietary formula through a reinforcement learning algorithm; the dietary formula is used to regulate energy metabolism.

[0066] Among them, personalized and dynamic feeding strategy optimization is carried out through a reinforcement learning algorithm: it can be based on the current metabolic state of the dairy cow, such as NEFA concentration, BHB threshold, environmental variables, and dietary components, to adjust the energy concentration of the diet, the dosage of added fat or vitamins, and modify the fiber ratio, etc. Based on the real-time feedback of milk production, health score such as the incidence of ketosis, cost control, and risk of nutritional excess such as digestive inhibition caused by excessive rumen bypass fat after adjusting the diet, iterative optimization strategy update is carried out. It can generate the optimal dietary formula through trial-and-error learning to achieve precise regulation of energy metabolism.

[0067] In one embodiment, the method further includes:

[0068] S201, extract multi-modal time-frequency domain features of perinatal basic information through time series coding and high-dimensional spectral analysis technology, and construct a three-dimensional spatio-temporal feature matrix based on the multi-modal time-frequency domain features; the perinatal basic information includes environmental parameters, continuous physiological signals, and metabolite indicators;

[0069] S202, based on the three-dimensional spatio-temporal feature matrix, construct a metabolic state prediction model using transfer learning technology;

[0070] S203, use the metabolic state prediction model to generate adjustment information, and update the feeding rule library based on the adjustment information.

[0071] Exemplarily, perinatal basic information includes environmental parameters such as temperature, humidity, etc., continuous physiological signals such as heart rate, blood pressure, etc., and metabolite indicators such as blood glucose, blood lipid, etc. By using time series coding and high-dimensional spectral analysis techniques, multi-modal time series frequency domain features of perinatal basic information are extracted. Time series coding can capture the variation law of data over time, and high-dimensional spectral analysis can map the data into a high-dimensional space to discover more subtle time series frequency domain features. Based on the extracted multi-modal time series frequency domain features, a three-dimensional spatio-temporal feature matrix is constructed. This matrix integrates different types of feature information in the spatial dimension and reflects the change of features over time in the time dimension, providing a comprehensive data basis for subsequent model construction.

[0072] Using the constructed three-dimensional spatio-temporal feature matrix, a model that can accurately predict the metabolic state of dairy cows is constructed through transfer learning technology. Transfer learning technology accelerates the training of the model and improves its performance by transferring existing knowledge and model parameters from one or more related fields to the current task. It can be to initialize the model parameters using the cow parturition dynamics simulation dataset, and a three-dimensional convolutional neural network is used to fuse time series-frequency domain features to construct a metabolic state prediction model. The metabolic risk probability is output through clinical detection indicators and the Sigmoid activation function (a common S-shaped function in biology, also known as the S-shaped growth curve), and the model is fine-tuned according to the actual data to obtain the metabolic state prediction model. Using the metabolic state prediction model, the metabolic state of dairy cows is predicted, and corresponding adjustment information is generated according to the prediction results. The adjustment information can include suggestions for adjusting the diet formula, increasing or decreasing nutrient components, etc. Based on the generated adjustment information, the feeding rule library is updated. The feeding rule library contains diet formula rules formulated according to the physiological state, metabolic requirements, etc. of dairy cows. By continuously updating and improving, it can better meet the special nutritional needs of dairy cows during the perinatal period and improve their production performance and health level.

[0073] In one embodiment, constructing a three-dimensional spatio-temporal feature matrix based on multi-modal time series frequency domain features includes:

[0074] S301, constructing a three-dimensional spatio-temporal feature matrix through the following formula:

[0075]

[0076] where τ represents the feature fusion function, LSTM represents the time series feature, FFT represents the frequency domain feature, P(t) represents the physiological signal time series data, M(t) represents the metabolite frequency domain feature, and h p / h m / h c represents the feature dimension of each modality.

[0077] Specifically, the τ feature fusion function is used to fuse features from different modalities such as physiological signals and metabolite indicators, effectively integrating features from different sources and of different natures, so that the fused features can more comprehensively and accurately reflect the comprehensive state of dairy cows during the peripartum period. LSTM can capture long-term dependencies in time-series data, such as the correlations and change trends between physiological signals of dairy cows at different time points, and obtain feature vectors that can reflect the evolution of physiological states over time. FFT is used to extract frequency-domain features of metabolite indicators. The physiological signal time-series data P(t) is the original data input into LSTM, representing the physiological signals continuously monitored for dairy cows during the peripartum period. The metabolite frequency-domain features M(t) are obtained by converting data of metabolite indicators such as blood glucose and blood lipids from the time domain to the frequency domain. The feature dimensions h p / h m / h c represent the feature dimensions of different modalities such as the physiological signal modality and the metabolite modality, etc. The time-series features extracted by LSTM and the frequency-domain features extracted by FFT are fused through the feature fusion function, and combined with the information of the feature dimensions of each modality, finally a three-dimensional spatio-temporal feature matrix is constructed. It can comprehensively characterize the physiological and metabolic characteristics of dairy cows during the peripartum period, providing a structured and information-rich feature representation for subsequent tasks such as metabolic state prediction.

[0078] In one embodiment, a metabolic state prediction model is constructed using transfer learning technology, including:

[0079] S401, initializing the model parameters using the cow parturition dynamics simulation dataset;

[0080] S402, using a three-dimensional convolutional neural network to fuse time-series - frequency-domain features, and outputting the metabolic risk probability and fine-tuning the model using the following formula:

[0081] P(y=1|T(t))=σ(W f ·F final (t)+W c ·C clinical )

[0082] where, F final represents the fused feature vector, C clinical represents the clinical detection index, and σ represents the Sigmoid activation function.

[0083] Exemplarily, in transfer learning, the model parameters are initialized using the cow parturition dynamics simulation dataset. This dataset is simulated and generated based on the physiological and metabolic changes during the cow parturition process, and contains a large number of samples and corresponding metabolic state labels and other information. Pre-training the model based on this dataset can make the initial state of the model parameters more reasonable. The fused feature vector F final, consisting of the feature vectors that fuse temporal - frequency domain features extracted by 3D convolution and clinical detection indicators, integrating multi - source information such as temporal, frequency domain, and clinical detection in the comprehensive multi - modal temporal - frequency domain feature matrix. Clinical detection indicator C clinical Indicators obtained from actual clinical detections, such as the concentration of non - esterified fatty acids in blood, the concentration of urinary urea nitrogen, etc. These indicators directly reflect the metabolic status of dairy cows and are important reference information for model prediction. The sigmoid activation function maps the fused feature vectors to the interval [0, 1], and the output value can be interpreted as the probability of the occurrence of metabolic risk. When the output value is greater than 0.5, it can be predicted that the dairy cow is in a high - metabolic - risk state; when the output value is less than or equal to 0.5, the dairy cow is in a low - risk state. This mapping method makes the output of the model have a clear probabilistic meaning, facilitating subsequent decision - making and analysis.

[0084] In one of the embodiments, the method further includes:

[0085] S501, calculating the dynamic energy metabolism index according to the obtained perinatal basic information;

[0086] S502, updating the feeding rule base according to the dynamic energy metabolism index, where the dynamic energy metabolism index is dynamically calculated by the following formula:

[0087]

[0088] where C BHB represents the concentration of plasma β - hydroxybutyric acid, C Trp represents the concentration of plasma tryptophan, CFU Cellulolytic represents the colony - forming unit of cellulose - degrading bacteria, CaO represents the calcium oxide content in the diet, CaD represents the total calcium content in the diet, NEFA represents the concentration of non - esterified fatty acids, and MUN represents the concentration of urinary urea nitrogen.

[0089] Specifically, the dynamic energy metabolism index is a comprehensive index used to quantitatively evaluate the energy metabolism level of dairy cows during the perinatal period, determine whether the dairy cows are in an energy - balanced state, and whether they are at risk of energy metabolism disorders. Using the dynamic energy metabolism index to update the corresponding diet adjustment strategies in the rule base, during subsequent feeding, the diet formula can be adjusted according to the actual energy metabolism status of dairy cows to meet the nutritional needs of dairy cows and maintain their health and production performance.

[0090] The concentration of BHB C BHB : The concentration of plasma β - hydroxybutyric acid in blood. BHB is one of the important indicators of energy metabolism in dairy cows, and its concentration change can reflect the energy state and metabolism of dairy cows. When there is energy deficiency or metabolic disorder, the BHB concentration may increase. The concentration of plasma tryptophan C Trp: The concentration of tryptophan in plasma. Tryptophan is an essential amino acid that participates in various physiological metabolic processes. Its concentration change is related to the nutritional status and metabolic activities of dairy cows and has a certain impact on energy metabolism. Colony-forming unit CFU of cellulose-degrading bacteria Cellulolytic : It represents the number of cellulose-degrading bacteria in the rumen of dairy cows, measured by colony-forming units. Cellulose-degrading bacteria can decompose cellulose during the digestion process of dairy cows, providing an energy source for dairy cows. The amount of it reflects the digestion and utilization ability of dairy cows for cellulose. Calcium oxide content CaO in the diet: The content of calcium oxide in the diet. Calcium oxide plays a certain role in the nutritional metabolism of dairy cows, affecting the absorption and utilization of calcium. Calcium participates in the activity regulation of various enzymes during the energy metabolism process of dairy cows. Therefore, the calcium oxide content in the diet has a certain impact on energy metabolism. Total calcium content CaD in the diet: The total calcium content in the diet. Calcium participates in various physiological functions such as bone formation, muscle contraction, and nerve conduction, and is also closely related to energy metabolism. Whether the total calcium content in the diet is appropriate affects the energy metabolism balance of dairy cows. Non-esterified fatty acid concentration NEFA: The concentration of non-esterified fatty acids in the blood. NEFA is an indicator of fat mobilization in dairy cows, and the change in its concentration can reflect the energy reserve and mobilization situation of dairy cows. Urinary urea nitrogen concentration MUN: The concentration of urea nitrogen in urine. MUN is closely related to the protein metabolism of dairy cows, and the change in its concentration can reflect the decomposition and synthesis of proteins.

[0091] In one of the embodiments, the method further includes:

[0092] S601, optimizing the weight parameters of the dynamic energy metabolism index formula using the following formula:

[0093]

[0094] where J(θ) represents the weight parameter, feed represents the feed formula parameter, y represents the production performance and health index, λ represents the rule stability penalty coefficient, ||θ new -θ old || 2 represents the Euclidean distance quantifying the difference between the old and new parameters.

[0095] Exemplarily, the weight parameter J(θ): In the calculation formula of the dynamic energy metabolism index, the influence degrees of various factors such as BHB concentration, plasma tryptophan concentration, etc. on the metabolism index are different. The weight parameter is used to measure the relative importance of each factor in the calculation. Feed formulation parameter feed: Various nutrient content, proportion and other parameters in the diet formulation, such as energy concentration, protein content, fat content, etc. These parameters directly affect the energy metabolism of dairy cows. Production performance and health index y: It refers to measuring the rationality and effectiveness of the diet formulation through the actual production performance of dairy cows such as milk yield, milk composition, etc. and health indexes such as blood biochemical indexes, disease incidence rate, etc. Rule stability penalty coefficient λ: A penalty coefficient is introduced to constrain the parameter change, ensure the stability of the food supply rule base, and avoid the instability of the rule caused by the frequent and large-scale adjustment of the weight parameter. The Euclidean distance ||θ new -θ old || 2 : It is used to measure the difference degree between the new and old weight parameters, and quantifies this difference through the Euclidean distance to ensure that the parameter update is within a reasonable range. This formula is used to optimize the weight parameter in the calculation of the dynamic energy metabolism index, so that the calculated metabolism index can more accurately reflect the actual energy metabolism state of dairy cows, and can be dynamically adjusted according to the production performance and health status of dairy cows.

[0096] In one embodiment, the method further includes:

[0097] S701, adjust the diet formulation according to the following steps:

[0098] According to the BHB and parity information of prepartum cows, adjust the dietary energy concentration in the diet formulation in stages, and add dry matter with a content of vitamin D3 2000 - 3000 IU / kg and 0.3% - 0.5% magnesium;

[0099] According to the results of rumen fluid metagenomic analysis, adjust the ratio of dietary fiber to starch in the diet formulation;

[0100] Dynamically set the BHB threshold according to parity, and add bypass fat to the diet formulation corresponding to cows with BHB exceeding the threshold within 7 days after calving.

[0101] Specifically, according to the BHB concentration and parity of dairy cows, the prepartum period is divided into different stages, and each stage corresponds to specific energy concentration requirements. For example, first-calving cows may require a higher energy concentration at a certain stage, while multiparous cows may require a lower energy concentration. Vitamin D3 helps with calcium absorption and utilization, and magnesium is involved in the regulation of the activity of various enzymes. Adding an appropriate amount of vitamin D3 at 2000 - 3000 IU / kg and 0.3% - 0.5% magnesium in dry matter can improve the energy metabolism and overall health of dairy cows. By analyzing the microbial community composition and functional genes in rumen fluid, the microbial metabolism in the rumen of dairy cows can be understood. Different microbial communities have different abilities to degrade fiber and starch. Therefore, the fiber and starch ratios in the diet can be adjusted according to the analysis results to optimize the fermentation efficiency of rumen microorganisms. Metagenomic analysis of rumen fluid can show the activities of fiber-degrading and starch-degrading microbial communities. If the activity of the fiber-degrading microbial community is high, the fiber ratio in the diet can be appropriately increased; if the activity of the starch-degrading microbial community is high, the starch ratio can be increased. Ensure that the nutritional components in the diet can be effectively utilized by rumen microorganisms to improve the energy conversion efficiency. The metabolic status and changes in BHB levels of dairy cows with different parities are different within 7 days after parturition. For example, the BHB thresholds of first-calving cows and multiparous cows may be different. By setting a dynamic BHB threshold, it is possible to determine whether a dairy cow is in a state of energy metabolism disorder. Cows with BHB levels exceeding the threshold may have problems with energy metabolism and require additional energy supplementation. Protected fat is a fat source that can stably exist in the rumen and be absorbed and utilized in the small intestine, providing additional energy for dairy cows without affecting the normal fermentation of rumen microorganisms. Adding protected fat can improve the energy status of dairy cows and reduce the risk of metabolic diseases.

[0102] In one embodiment, the method further includes:

[0103] S801, adjusting the diet formula of postpartum dairy cows according to the following stages:

[0104] In the first week after parturition, the dietary energy supply concentration is adjusted to: 1.45 - 1.50 Mcal / kg DM;

[0105] In the second week after parturition, the dietary energy supply concentration is adjusted to: 1.51 - 1.55 Mcal / kg DM;

[0106] From the third week after parturition to the end of the lactation peak, the dietary energy supply concentration is adjusted to: 1.56 - 1.60 Mcal / kg DM.

[0107] Exemplarily, in the first week after calving, the dry matter intake of dairy cows is low, but the energy requirement increases sharply. At this time, a diet with a higher energy density is needed to meet the basic needs of dairy cows while avoiding digestive problems caused by excessive energy supply. In the second week after calving, as dairy cows gradually adapt to the physiological changes during the lactation period, the dry matter intake begins to increase but is still insufficient to meet the high energy requirements. At this time, appropriately increasing the energy concentration of the diet can further support the milk production and energy balance of dairy cows. From the 3rd week after calving to the end of the lactation peak, the milk production of dairy cows reaches the peak, and the energy requirement also reaches the highest. At this time, further increasing the energy concentration of the diet can ensure that dairy cows obtain sufficient energy to support high milk production.

[0108] In one of the embodiments, the method further includes:

[0109] S901, adjusting the diet formula of dairy cows during the lactation peak according to the following rules:

[0110] Maintain the dietary calcium concentration in the range of 0.90% - 1.10% on a dry matter basis;

[0111] When the dietary energy supply concentration ≥ 1.55 Mcal / kg DM, add 250 - 300 g / d of bypass fat;

[0112] When it is detected that plasma NEFA > 1.2 mmol / L, add 500 IU / d of vitamin E;

[0113] When it is detected that MUN > 15 mg / dL, reduce the dietary crude protein to 16% - 17%.

[0114] Specifically, during the peak lactation period, the calcium demand of dairy cows increases. If the calcium content in the diet is insufficient, it may lead to hypocalcemia in dairy cows, affecting their health and production performance. By controlling the calcium concentration in the diet within the range of 0.90%-1.10% on a dry matter basis, appropriate calcium intake for dairy cows can be ensured. During the peak lactation period, when the dietary energy supply concentration reaches or exceeds 1.55 Mcal / kg DM, dairy cows may still require additional energy sources. Rumen-protected fat can provide additional energy without affecting the normal fermentation of rumen microorganisms. Increasing rumen-protected fat can meet the additional energy requirements of dairy cows. Regularly detect the NEFA concentration in the plasma of dairy cows. If the NEFA concentration exceeds 1.2 mmol / L, it indicates that the dairy cows may be in a state of negative energy balance. Adding 500 IU of vitamin E can enhance the immunity and antioxidant capacity of dairy cows and reduce the impact brought by the negative energy balance state. Milk urea nitrogen (MUN) is an indicator of protein metabolism in dairy cows. An increase in its concentration usually indicates that the crude protein content in the diet is too high or the protein utilization rate is low. Excessive crude protein content not only increases the burden on the liver and kidneys of dairy cows but also may lead to an increase in nitrogen emissions. The MUN concentration in the urine of dairy cows can be regularly detected. If the MUN concentration exceeds 15 mg / dL, it indicates that the crude protein content in the diet may be too high. Reduce the crude protein content in the diet to 16%-17%. This can be achieved by reducing the usage of high-protein feeds such as soybean meal and fish meal and increasing the usage of low-protein feeds such as corn and wheat. Through the adjustment of the above rules, the dietary formula of dairy cows during the peak lactation period can be effectively optimized to meet their high-energy and high-nutrition requirements, improving production performance and health level.

[0115] A method for regulating energy metabolism in the periparturient period of dairy cows in this application realizes the precise regulation of energy metabolism in the periparturient period of dairy cows through multimodal data fusion and intelligent algorithms. Its innovation lies in comprehensively integrating multi-source data such as environmental parameters, continuous physiological signals, and metabolite indicators, constructing a three-dimensional spatio-temporal feature matrix, generating a multimodal knowledge graph, and deeply mining the deep features and complex relationships in the data. At the same time, using transfer learning technology and reinforcement learning algorithms, a metabolic state prediction model is constructed to dynamically update the feeding rule base and generate personalized dietary formulas. This method can not only reflect the energy metabolism state of dairy cows in real time but also optimize the weight parameters of the dynamic energy metabolism index formula according to actual needs, further improving the accuracy of regulation. At different physiological stages, the key parameters in the dietary formula can be adjusted in stages to meet the nutritional needs of dairy cows, reduce the occurrence of metabolic diseases, and improve the production performance of dairy cows. In addition, the integrated automation design of this invention enables the efficient realization of the entire process from data collection to dietary formula generation, which can reduce breeding costs and environmental pollution, and has significant economic and social benefits.

[0116] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are sequentially shown according to the indications of the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, there is no strict order limit for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same moment, but can be executed at different moments. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0117] Based on the same inventive concept, an embodiment of the present application further provides a system for implementing a method for regulating energy metabolism during the peripartum period of dairy cows as described above. The solution provided by this system to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of a dairy cow peripartum energy metabolism regulation system provided below can refer to the limitations on a dairy cow peripartum energy metabolism regulation method in the above text, and will not be repeated here.

[0118] In an exemplary embodiment, as Figure 2 shown, a dairy cow peripartum energy metabolism regulation system is provided, including a peripartum information processing module 11, a peripartum atlas generation module 12, a feeding rule library generation module 13, and a daily ration formula generation module 14:

[0119] The peripartum information processing module 11 is used to obtain peripartum basic information and perform structured processing to obtain peripartum data information;

[0120] The peripartum atlas generation module 12 is used to extract the association features between entities in the peripartum data information through a graph neural network to generate a multimodal knowledge graph;

[0121] The feeding rule library generation module 13 is used to combine historical metabolic data and domain expert experience, and use a deep belief network to perform representation learning on the association features in the multimodal knowledge graph to generate a feeding rule library;

[0122] The daily ration formula generation module 14 is used to generate a daily ration formula based on the feeding rule library through a reinforcement learning algorithm; the daily ration formula is used to regulate energy metabolism.

[0123] In an embodiment, the system further includes a rule library update module for:

[0124] Extract the multi-modal time-frequency domain features of perinatal basic information through time series coding and high-dimensional spectral analysis techniques, and construct a three-dimensional spatio-temporal feature matrix based on the multi-modal time-frequency domain features; the perinatal basic information includes environmental parameters, continuous physiological signals, and metabolite indicators;

[0125] Based on the three-dimensional spatio-temporal feature matrix, use transfer learning technology to construct a metabolic state prediction model;

[0126] Use the metabolic state prediction model to generate adjustment information, and update the feeding rule base based on the adjustment information.

[0127] In one embodiment, the rule base update module is further configured to:

[0128] Construct a three-dimensional spatio-temporal feature matrix through the following formula:

[0129]

[0130] where τ represents the feature fusion function, LSTM represents the time series feature, FFT represents the frequency domain feature, P(t) represents the physiological signal time series data, M(t) represents the metabolite frequency domain feature, h p / h m / h c represents the feature dimensions of each modality.

[0131] In one embodiment, the rule base update module is further configured to:

[0132] Initialize the model parameters using the cow parturition dynamics simulation data set;

[0133] Use a three-dimensional convolutional neural network to fuse the time series-frequency domain features, and use the following formula to output the metabolic risk probability and fine-tune the model:

[0134] P(y=1|T(t))=σ(W f ·F final (t)+W c ·C clinical )

[0135] where F final represents the fused feature vector, C clinical represents the clinical detection index, and σ represents the Sigmoid activation function.

[0136] In one embodiment, the rule base update module is further configured to:

[0137] Calculate the dynamic energy metabolism index according to the obtained perinatal basic information;

[0138] Update the feeding rule base according to the dynamic energy metabolism index, where the dynamic energy metabolism index is dynamically calculated through the following formula:

[0139]

[0140] Among them, C BHB represents the plasma β-hydroxybutyric acid concentration, C Trp represents the plasma tryptophan concentration, CFU Cellulolytic represents the colony forming unit of cellulose-degrading bacteria, CaO represents the dietary calcium oxide content, CaD represents the total dietary calcium content, NEFA represents the non-esterified fatty acid concentration, and MUN represents the urinary urea nitrogen concentration.

[0141] In one embodiment, the rule base update module is further configured to:

[0142] Optimize the weight parameters of the dynamic energy metabolism index formula using the following formula:

[0143]

[0144] Among them, J(θ) represents the weight parameter, feed represents the feed formulation parameter, y represents the production performance and health index, λ represents the rule stability penalty coefficient, ||θ new -θ old || 2 represents the Euclidean distance quantifying the difference between the old and new parameters.

[0145] In one of the embodiments, the dietary formulation generation module 14 further includes a custom diet unit for:

[0146] Adjust the dietary formulation according to the following steps:

[0147] According to the BHB and parity information of prepartum cows, adjust the dietary energy concentration in the dietary formulation in stages, and add dry matter with a vitamin D3 content of 2000 - 3000 IU / kg and 0.3% - 0.5% magnesium;

[0148] According to the results of rumen fluid metagenomic analysis, adjust the ratio of dietary fiber to starch in the dietary formulation;

[0149] Dynamically set the BHB threshold according to parity, and add bypass fat to the dietary formulation corresponding to cows with BHB above the threshold within 7 days after calving.

[0150] In one of the embodiments, the custom diet unit is further used for:

[0151] Adjust the dietary formulation of postpartum cows according to the following stages:

[0152] In the first week after calving, adjust the dietary energy supply concentration to: 1.45 - 1.50 Mcal / kg DM;

[0153] In the second week after parturition, the dietary energy supply concentration is adjusted to: 1.51 - 1.55 Mcal / kg DM;

[0154] From the third week after parturition to the end of the lactation peak period, the dietary energy supply concentration is adjusted to: 1.56 - 1.60 Mcal / kg DM.

[0155] In one embodiment, the customized dietary unit is further used for:

[0156] Adjusting the dietary formula of dairy cows at the lactation peak according to the following rules:

[0157] Maintaining the dietary calcium concentration in the range of 0.90% - 1.10% on a dry matter basis;

[0158] When the dietary energy supply concentration ≥ 1.55 Mcal / kg DM, add 250 - 300 g / d of bypass fat;

[0159] When it is detected that plasma NEFA > 1.2 mmol / L, add 500 IU / d of vitamin E;

[0160] When it is detected that MUN > 15 mg / dL, reduce the dietary crude protein to 16% - 17%.

[0161] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of a method for regulating energy metabolism during the periparturient period of dairy cows as described above are implemented.

[0162] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0163] For the device embodiment, since it basically corresponds to the method embodiment, the relevant parts can be referred to the partial description of the method embodiment. The device embodiments described above are only illustrative. The components described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present disclosure solution. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0164] The above-described embodiments merely represent several implementation manners of the embodiments of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the embodiments of the application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the embodiments of the present application, several variations and improvements can still be made, and these all fall within the protection scope of the embodiments of the present application.

Claims

1. A method for regulating energy metabolism during the periparturient period of dairy cows, characterized in that, The method includes: Obtaining perinatal basic information, performing structured processing, and obtaining perinatal data information; Extracting the association features between entities in the perinatal data information through a graph neural network to generate a multimodal knowledge graph; Combining historical metabolic data and domain expert experience, and using a deep belief network to perform representation learning on the association features in the multimodal knowledge graph to generate a feeding rule base; Based on the feeding rule base, generating a daily ration formula through a reinforcement learning algorithm; the daily ration formula is used to regulate energy metabolism.

2. The method according to claim 1, characterized in that, The method further includes: Extracting the multimodal time-frequency domain features of the perinatal basic information through time series encoding and high-dimensional spectral analysis technology, and constructing a three-dimensional spatio-temporal feature matrix based on the multimodal time-frequency domain features; the perinatal basic information includes environmental parameters, continuous physiological signals, and metabolite indicators; Based on the three-dimensional spatio-temporal feature matrix, constructing a metabolic state prediction model using transfer learning technology; Using the metabolic state prediction model to generate adjustment information, and updating the feeding rule base based on the adjustment information.

3. The method according to claim 2, wherein The constructing of the three-dimensional spatio-temporal feature matrix based on the multimodal time-frequency domain features includes: Constructing the three-dimensional spatio-temporal feature matrix through the following formula: Among them, τ represents the feature fusion function, LSTM represents the temporal feature, FFT represents the frequency domain feature, P(t) represents the physiological signal temporal data, M(t) represents the metabolite frequency domain feature, and h p / h m / h c represents the dimensionality of each modal feature.

4. The method according to claim 3, wherein The constructing of the metabolic state prediction model using transfer learning technology includes: Initializing the model parameters using a cow parturition dynamics simulation data set; Using a three-dimensional convolutional neural network to fuse time-frequency domain features, and using the following formula to output the metabolic risk probability and perform model fine-tuning: P(y = 1|T(t)) = σ(W f ·F final (t) + W c ·C clinical ) Among them, F final represents the fusion feature vector, C clinical represents the clinical detection index, and σ represents the Sigmoid activation function.

5. The method according to claim 1, characterized in that, The method further includes: Calculating a dynamic energy metabolism index according to the obtained perinatal basic information; Updating the feeding rule base according to the dynamic energy metabolism index, where the dynamic energy metabolism index is dynamically calculated through the following formula: Among them, C BHB represents the plasma β-hydroxybutyric acid concentration, C Trp represents the plasma tryptophan concentration, CFU Cellulolytic represents the colony-forming unit of cellulose-degrading bacteria, CaO represents the dietary calcium oxide content, CaD represents the total dietary calcium content, NEFA represents the non-esterified fatty acid concentration, and MUN represents the urinary urea nitrogen concentration.

6. The method according to claim 5, characterized in that, The method further includes: Optimizing the weight parameters of the dynamic energy metabolism index formula using the following formula: Among them, J(θ) represents the weight parameter, feed represents the feed formulation parameter, y represents the production performance and health index, λ represents the rule stability penalty coefficient, ||θ new - θ old || 2 represents the Euclidean distance quantifying the difference between the old and new parameters.

7. The method according to claim 1, characterized in that The method further includes: Adjusting the daily ration formula according to the following steps: According to the BHB and parity information of cows in the prepartum period, adjusting the energy concentration of the daily ration in the daily ration formula in stages, and adding dry matter with a vitamin D3 content of 2000 - 3000 IU / kg and 0.3% - 0.5% magnesium; Adjusting the ratio of dietary fiber to starch in the daily ration formula according to the results of rumen fluid metagenomic analysis; Dynamically setting the BHB threshold according to parity, and adding bypass fat to the daily ration formula corresponding to cows with a BHB exceeding the threshold within 7 days after parturition.

8. The method according to claim 6, characterized in that The method further includes: Adjusting the daily ration formula of postpartum cows according to the following stages: In the first week after parturition, the dietary energy supply concentration is adjusted to: 1.45 - 1.50 Mcal / kg DM; In the second week after parturition, the dietary energy supply concentration is adjusted to: 1.51 - 1.55 Mcal / kg DM; From the third week after parturition to the end of the lactation peak period, the dietary energy supply concentration is adjusted to: 1.56 - 1.60 Mcal / kg DM.

9. The method according to claim 1, wherein The method further includes: Adjusting the daily ration formula of cows in the lactation peak period according to the following rules: Maintaining the dietary calcium concentration in the range of 0.90% - 1.10% on a dry matter basis; When the dietary energy supply concentration ≥ 1.55 Mcal / kg DM, add 250 - 300 g / d of rumen-protected fat; When plasma NEFA > 1.2 mmol / L is detected, add 500 IU / d of vitamin E; When MUN > 15 mg / dL is detected, reduce the dietary crude protein to 16% - 17%.

10. A periparturient energy metabolism regulation system for dairy cows, characterized in that, The system includes a periparturient information processing module, a periparturient atlas generation module, a feeding rule library generation module, and a dietary formula generation module: The periparturient information processing module is used to obtain periparturient basic information and perform structured processing to obtain periparturient data information; The periparturient atlas generation module is used to extract the association features between entities in the periparturient data information through a graph neural network to generate a multimodal knowledge graph; The feeding rule library generation module is used to combine historical metabolic data and domain expert experience, and use a deep belief network to perform representation learning on the association features in the multimodal knowledge graph to generate a feeding rule library; The dietary formula generation module is used to generate a dietary formula based on the feeding rule library through a reinforcement learning algorithm.

Images