Silage data processing device and method for predicting rumen digestion gas production and computer program

Abstract

The invention discloses a silage data processing device, method and computer program for predicting rumen digestion gas production. Based on silage properties (13 input variables) and rumen gas production and digestion results (14 output variables at different moments), a model for predicting the gas production and digestion results of the silage in the rumen is constructed by using five tree models such as CatBoost and the like. R2 of the total gas production rate prediction model and R2 of the methane yield prediction model can be higher than 0.83, and the highest R2 can reach 0.96; the R2 of the digestibility model is more than 0.79, except for the IVDMD24h model (0.58), the data processing device and method can be applied to precise feeding and green breeding of ruminants.

Description

Technical Field

[0001] This invention belongs to the field of biotechnology, and specifically relates to a data processing device, method and computer program for predicting silage digestion and gas production. Background Technology

[0002] Against the backdrop of global climate change, greenhouse gas emissions, particularly methane (CH4) and carbon dioxide (CO2), remain key contributors to rising temperatures. It is estimated that livestock production accounts for 14.5% of total anthropogenic greenhouse gas emissions (PJ Gerber et al., Tackling Climate Change through Livestock: A Global Assessment of Emissions and Mitigation Opportunities, PJ Gerber et al., Eds. (Food and Agriculture Organization of the United Nations–FAO, Rome, Italy, 2013)), with approximately 80% of greenhouse gas emissions and 90% of methane emissions originating from ruminants (Schultz MM, Neser FWC, Makgahlela ML. A balanced perspective on the importance of extensive ruminant production for human nutrition and livelihoods and its contribution to greenhouse gas emissions[J]. South African Journal of Science, 2020, 116(9-10): 1-3).

[0003] In ruminants, the rumen produces energy and a large amount of gas during digestion as various microorganisms break down organic matter in feed. These gases primarily consist of carbon dioxide, methane, and hydrogen. Methanogenic archaea play a crucial role, converting CO2 and hydrogen (H2) into CH4. This process not only consumes the animal's energy but is also harmful to the environment. Therefore, in order to promote green farming and mitigate global warming, more and more researchers are exploring how to formulate feed that produces less gas while still providing the energy needed for animal growth.

[0004] Rumen digestibility is a core indicator of feed nutrient conversion efficiency—high digestibility means more energy and nutrients are absorbed by the animal for production activities such as muscle growth and milk synthesis, and is a key basis for evaluating feed nutritional value and animal production performance. In vitro rumen digestion experiments are the most commonly used method to assess the impact of feed on rumen gas production and digestibility. The general procedure involves mixing rumen fluid and trace element solutions in a specific ratio to form an in vitro fermentation broth, then placing the feed into the broth for fermentation for a certain period, and using instruments to measure gas production. Compared to in vivo digestion experiments, this method is simpler, less costly, and causes less harm to the animal.

[0005] Silage is a widely used and efficient feed processing method that uses anaerobic fermentation technology to seal and store fresh forage or crop raw materials, utilizing lactic acid bacteria fermentation to inhibit spoilage and preserve nutrients for a long time. Its soft texture and tangy aroma can increase feed intake in ruminants, making it an important forage preparation technology in modern animal husbandry. The fermentation quality of silage has a significant impact on rumen fermentation. For example, the fermentation quality of corn silage affects rumen fermentation, methane emissions, and plasma metabolites in dairy cows. The pH of the feed affects the activity of rumen microorganisms and fermentation efficiency. Lactic acid and acetic acid in the rumen help maintain pH and inhibit the growth of harmful bacteria, thereby improving fermentation quality. Simultaneously, when acetic acid is produced in large quantities, it increases the hydrogen concentration in the rumen, thereby stimulating methanogenic bacteria activity and leading to increased methane production. An increase in propionic acid can significantly reduce the number and activity of methanogenic bacteria because excess hydrogen is used for propionic acid synthesis, thus reducing the hydrogen available for methane synthesis.

[0006] The chemical composition of feed also affects rumen fermentation, including factors such as detergent fiber, neutral detergent fiber, protein, and ash. Detergent fiber and neutral detergent fiber are important indicators of feed nutritional value. Feeds high in detergent fiber and neutral detergent fiber typically have lower energy density but provide a longer feeling of fullness, helping to control weight and maintain rumen health. However, high-fiber forages produce more gas. High-fiber diets often lead to acetic acid fermentation, producing more greenhouse gases. Protein is the primary nutrient source for the growth and reproduction of rumen microorganisms. High-protein feeds can promote the activity of beneficial microorganisms, thereby improving fermentation quality and reducing ammonia nitrogen production. Minerals in ash, such as calcium and phosphorus, are crucial for maintaining the normal function of rumen microorganisms. Appropriate mineral content helps regulate pH and optimize the microbial environment. Propionic acid is also produced during carbohydrate breakdown; as a hydrogen acceptor, it maximizes the utilization of hydrogen produced by acetic acid fermentation, thereby reducing methane production. In summary, there is a close relationship between the quality of silage (including its chemical composition and fermentation state) and rumen gas production. Improving the quality of silage can effectively regulate the fermentation process in the rumen, thereby affecting methane emissions and the production of other greenhouse gases. However, the specific indicators of silage quality and rumen gas production have a complex relationship, and there is currently no model or equation that can explain this relationship. Summary of the Invention

[0007] The technical problem to be solved by this invention is how to detect, predict, or assist in predicting the greenhouse gas emissions (potential) of silage in the rumen of ruminants and / or how to predict or assist in predicting the greenhouse gas emissions of silage in the rumen of ruminants without relying on in vitro rumen experiments and / or how to achieve precise feeding of ruminants.

[0008] To address the aforementioned technical problems, the present invention first provides a data processing apparatus, which may include a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to perform the following steps: A1) Model Construction: Based on 13 model input variables and model output variables, a machine learning algorithm was used to construct a model to predict the gas production and digestion of silage in the rumen of ruminants. The 13 model input variables are the pH of the silage, the lactic acid (LA) content, acetic acid (AA) content, propionic acid (PA) content, butyric acid (BA) content, ammonia nitrogen (NH3-N) content, dry matter (DM) content, soluble carbohydrate (WSC) content, crude protein (CP) content, crude fat (EE) content, neutral detergent fiber (NDF) content, acid detergent fiber (ADF) content, and crude ash (Ash) content. The model output variables are the gas production and / or digestibility of the silage in the rumen. A2) Model prediction output: Input the 13 model input variables of the silage to be tested into the model obtained in A1), and output the gas production and digestion results of the silage to be tested in the rumen of ruminants, the gas production and digestion results including gas production and / or digestibility.

[0009] In one specific embodiment of the present invention, the model output variables are obtained based on in vitro rumen digestion experimental data of silage.

[0010] In the aforementioned data processing device, the gas production may include the total gas production and / or methane production at different times in the rumen; the digestibility may include the dry matter digestibility and / or detergent fiber digestibility at different times in the rumen; and the detergent fiber may include neutral detergent fiber and / or acid detergent fiber.

[0011] In the aforementioned data processing device, the dry matter may specifically be the substance remaining after drying off the moisture.

[0012] The total gas production may include 24-hour total gas production (total gas production during 24 hours of rumen digestion), 36-hour total gas production (total gas production during 36 hours of rumen digestion), 48-hour total gas production (total gas production during 48 hours of rumen digestion), or / and 72-hour total gas production (total gas production during 72 hours of rumen digestion).

[0013] The methane production may include 24h methane production (methane production after 24 hours of rumen digestion), 48h methane production (methane production after 48 hours of rumen digestion), or / and 72h methane production (methane production after 72 hours of rumen digestion).

[0014] The dry matter digestibility may include 24-hour dry matter digestibility (dry matter digestibility after 24 hours of rumen digestion), 48-hour dry matter digestibility (dry matter digestibility after 48 hours of rumen digestion), or / and 72-hour dry matter digestibility (dry matter digestibility after 72 hours of rumen digestion).

[0015] The digestibility of the detergent fiber may include 48h neutral detergent fiber digestibility (neutral detergent fiber digestibility after 48 hours of rumen digestion), 72h neutral detergent fiber digestibility (neutral detergent fiber digestibility after 72 hours of rumen digestion), 48h acid detergent fiber digestibility (acid detergent fiber digestibility after 48 hours of rumen digestion) or / and 72h acid detergent fiber digestibility (acid detergent fiber digestibility after 72 hours of rumen digestion).

[0016] In the aforementioned data processing device, the machine learning algorithm may be CatBoost.

[0017] To address the aforementioned technical problems, the present invention also provides a method for predicting the gas production and digestion of silage in the rumen, comprising: B1) Model Construction: Based on 13 model input variables and model output variables, a machine learning algorithm was used to construct a model to predict the gas production and digestion of silage in the rumen of ruminants. The 13 model input variables are the pH of the silage, the lactic acid (LA) content, acetic acid (AA) content, propionic acid (PA) content, butyric acid (BA) content, ammonia nitrogen (NH3-N) content, dry matter (DM) content, soluble carbohydrate (WSC) content, crude protein (CP) content, crude fat (EE) content, neutral detergent fiber (NDF) content, acid detergent fiber (ADF) content, and crude ash (Ash) content. The model output variables are the gas production and / or digestibility of the silage in the rumen. B2) Model prediction output: Input the 13 model input variables of the silage to be tested into the model obtained in B1), and output the gas production and digestion results of the silage to be tested in the rumen of ruminants, the gas production and digestion results including gas production and / or digestibility.

[0018] In one specific embodiment of the present invention, the model output variables are based on in vitro rumen digestion experimental data of silage.

[0019] In the above method, the gas production includes total gas production and / or methane production; the digestibility includes dry matter digestibility and / or detergent fiber digestibility; and the detergent fiber includes neutral detergent fiber and / or acid detergent fiber.

[0020] In the above method, the dry matter can specifically be the substance remaining after drying off the moisture.

[0021] The total gas production may include 24-hour total gas production (total gas production during 24 hours of rumen digestion), 36-hour total gas production (total gas production during 36 hours of rumen digestion), 48-hour total gas production (total gas production during 48 hours of rumen digestion), or / and 72-hour total gas production (total gas production during 72 hours of rumen digestion).

[0022] The methane production may include 24h methane production (methane production after 24 hours of rumen digestion), 48h methane production (methane production after 48 hours of rumen digestion), or / and 72h methane production (methane production after 72 hours of rumen digestion).

[0023] The dry matter digestibility may include 24-hour dry matter digestibility (dry matter digestibility after 24 hours of rumen digestion), 48-hour dry matter digestibility (dry matter digestibility after 48 hours of rumen digestion), or / and 72-hour dry matter digestibility (dry matter digestibility after 72 hours of rumen digestion).

[0024] The digestibility of the detergent fiber may include 48h neutral detergent fiber digestibility (neutral detergent fiber digestibility after 48 hours of rumen digestion), 72h neutral detergent fiber digestibility (neutral detergent fiber digestibility after 72 hours of rumen digestion), 48h acid detergent fiber digestibility (acid detergent fiber digestibility after 48 hours of rumen digestion) or / and 72h acid detergent fiber digestibility (acid detergent fiber digestibility after 72 hours of rumen digestion).

[0025] In the above method, the machine learning algorithm can be CatBoost.

[0026] To address the aforementioned technical problems, the present invention also provides an apparatus for predicting the gas production and digestion of silage in the rumen, the apparatus comprising the following modules: C1) Model Building Module: Used to construct a model predicting the gas production and digestibility of silage in the rumen of ruminants using machine learning algorithms, based on 13 model input variables and model output variables. The 13 model input variables are the pH of the silage, the lactic acid (LA) content, acetic acid (AA) content, propionic acid (PA) content, butyric acid (BA) content, ammonia nitrogen (NH3-N) content, dry matter (DM) content, soluble carbohydrate (WSC) content, crude protein (CP) content, crude fat (EE) content, neutral detergent fiber (NDF) content, acid detergent fiber (ADF) content, and crude ash (Ash) content. The model output variable is the gas production and / or digestibility of the silage in the rumen. C2) Model prediction output module: used to input the 13 model input variables of the silage to be tested into the model obtained in C1), and output the gas production and digestion results of the silage to be tested in the rumen of ruminants, the gas production and digestion results including gas production and / or digestibility.

[0027] In one specific embodiment of the present invention, the model output variable is obtained based on in vitro rumen digestion experiment data of a unit mass of silage.

[0028] In the above-mentioned apparatus, the gas production includes total gas production and / or methane production; the digestibility includes dry matter digestibility and / or detergent fiber digestibility; and the detergent fiber includes neutral detergent fiber and / or acid detergent fiber.

[0029] The dry matter can specifically be the substance remaining after drying off the moisture.

[0030] The total gas production may include 24-hour total gas production (total gas production during 24 hours of rumen digestion), 36-hour total gas production (total gas production during 36 hours of rumen digestion), 48-hour total gas production (total gas production during 48 hours of rumen digestion), or / and 72-hour total gas production (total gas production during 72 hours of rumen digestion).

[0031] The methane production may include 24h methane production (methane production after 24 hours of rumen digestion), 48h methane production (methane production after 48 hours of rumen digestion), or / and 72h methane production (methane production after 72 hours of rumen digestion).

[0032] The dry matter digestibility may include 24-hour dry matter digestibility (dry matter digestibility after 24 hours of rumen digestion), 48-hour dry matter digestibility (dry matter digestibility after 48 hours of rumen digestion), or / and 72-hour dry matter digestibility (dry matter digestibility after 72 hours of rumen digestion).

[0033] The digestibility of the detergent fiber may include 48h neutral detergent fiber digestibility (neutral detergent fiber digestibility after 48 hours of rumen digestion), 72h neutral detergent fiber digestibility (neutral detergent fiber digestibility after 72 hours of rumen digestion), 48h acid detergent fiber digestibility (acid detergent fiber digestibility after 48 hours of rumen digestion) or / and 72h acid detergent fiber digestibility (acid detergent fiber digestibility after 72 hours of rumen digestion).

[0034] The machine learning algorithm may be CatBoost.

[0035] To address the aforementioned technical problems, the present invention also provides a computer program product, comprising a computer program that, when executed by a processor, can implement the steps of the method described above.

[0036] To address the aforementioned technical problems, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, can implement the steps of the method described above.

[0037] The following applications of the data processing apparatus and / or the computer program product and / or the computer-readable storage medium described above also fall within the scope of protection of this invention: C1) Application in the preparation of green feed for ruminants; Application of C2 in the preparation of precision feeding products for ruminants.

[0038] The computer program products mentioned above can be software products that primarily implement their solutions through computer programs.

[0039] The computer-readable storage medium mentioned above refers to a carrier for storing data, which may be magnetic tape, disk, floppy disk, optical disk, magneto-optical disk, ROM, PROM, VCD, DVD, hard disk, flash memory, USB flash drive, CF card, SD card, MMC card, SM card, Memory Stick, or xD card, etc.

[0040] This invention aims to utilize interpretable machine learning modeling techniques to visualize the complex relationships between diverse silage fermentation properties and rumen gas production, methane production, and digestibility at multiple time points. By collecting and integrating experimental data from comprehensive literature on relevant topics, a tree model algorithm is used to construct a desired and accurate predictive model to quickly assess the rumen greenhouse gas emission potential and help silage researchers reduce the complexity of rumen experiments.

[0041] After feed is ingested into the rumen of ruminants, fermentation produces numerous greenhouse gases, a key factor in the "greenhouse effect." Fermented silage is a core component ensuring the year-round feed supply and nutritional needs of livestock. However, the complex relationship between silage and the rumen remains unclear. This invention uses interpretable machine learning techniques to link the intrinsic relationship between silage properties (13 input variables) and rumen gas production and digestion (14 output variables at different time points). Through comparison of various tree models, the R-values ​​of the total gas production prediction model and the methane production prediction model are compared. 2 All could be higher than 0.83, with the highest reaching 0.96; the R-value of the digestibility model 2 All values ​​were above 0.79, except for the IVCMD_24h model (0.58), with the CatBoost algorithm showing the best performance. Feature analysis showed that the quality of silage fermentation had a greater impact on the early stage of rumen fermentation (24h) than chemical composition, but the opposite was true in the middle and late stages of rumen fermentation (48h, 72h). This invention quantifies the specific impact of each feature on rumen gas production, methane production, and dry matter digestibility, which are all different, and the influence of variations in the specific feature value range is captured. Therefore, this invention is the first to analyze the complex relationship between silage and rumen gas production and digestion from a machine learning perspective. The quantified feature range influences and the website can help farmers and herders make decisions on how to feed silage precisely, achieve green farming, and mitigate global warming.

[0042] This invention innovatively utilizes machine learning to analyze the complex relationship between silage fermentation and rumen gas production, methane production, and digestibility, providing new insights into the field from a model perspective. Furthermore, the optimal model prediction method and products developed in this invention can replace traditional time-consuming and costly rumen experiments. This not only protects experimental animals to a certain extent but also guides farmers and herders to formulate feed in a targeted manner, thereby achieving the vision of precise feeding, cost reduction and efficiency improvement, and environmental protection. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the process of this invention.

[0044] Figure 2 The distribution of the collected experimental data.

[0045] Figure 3 This is a regression plot based on the optimal CatBoost model.

[0046] Figure 4 The feature importance of the optimal CatBoost model is graded.

[0047] Figure 5 SHAP analysis for the optimal CatBoost model.

[0048] Figure 6 The results of a one-dimensional partial dependency analysis of some features of the optimal CatBoost model.

[0049] Figure 7 The mean absolute error of the optimal model.

[0050] Figure 8 Information from 79 literature sources used to construct the model for this invention.

[0051] Figure 9 To verify the optimal model for each output variable of this invention, three additional experimental data were conducted to compare the predicted and actual values ​​of silage gas production or digestibility in the rumen with the actual values.

[0052] Figure 10 Three additional experimental data and literature information. Detailed Implementation

[0053] The present invention will now be described in further detail with reference to specific embodiments. The given embodiments are merely illustrative of the invention and not intended to limit its scope. The embodiments provided below can serve as a guide for further improvements by those skilled in the art and do not constitute a limitation on the invention in any way.

[0054] Unless otherwise specified, the experimental methods used in the following examples are conventional methods, performed according to the techniques or conditions described in the literature in this field or according to the product instructions. Unless otherwise specified, the materials and reagents used in the following examples are commercially available.

[0055] Example 1. Model Construction and Analysis for Predicting Greenhouse Gas Emissions from Silage in the Rumen of Ruminants 1. Data Analysis and Optimal Model Construction 1.1 Data Acquisition and Processing Using the following search terms: ((TS=("in vitro rumen")) OR TS=("in vitro gas")) AND TS=(silage), 525 papers containing complete data on silage fermentation quality and nutritional composition (i.e., the 13 model input variables described below) were collected from the Web of Science database (before March 11, 2025). After carefully reviewing each paper, 79 papers contained complete data on silage fermentation quality and nutritional composition. Figure 8 () is reserved.

[0056] The 79 papers included a variety of silage feedstocks, such as corn, alfalfa, oats, sorghum, and broccoli. This diverse data source increases the model's generalization ability, making it applicable not only to the analysis of a few specific types of silage. Successful silages are characterized by low acidity; the pH range in these diverse data points is 3.34–6.76, with a median of 4.3. It is well known that a pH of around 4.2 after silage fermentation indicates good fermentation; therefore, the experimental data show that different feedstocks can ferment well under appropriate treatments (e.g., adding inoculants and mixing silage). Among the organic acid indicators, lactic acid had the highest proportion, consistent with the general rule that silage is dominated by lactic acid. In terms of chemical composition, the dry matter data ranged from 66.6–934 g / kg FM, with a median of 346.2 g / kg, indicating a high moisture content in the diversified forage silage. Experimental data show that neutral detergent fiber (median 505.99 g / kg DM) is higher than acid detergent fiber (median 313 g / kg DM), because neutral detergent fiber contains hemicellulose compared to acid detergent fiber.

[0057] This invention collected a total of 13 model input variables, which were divided into two categories: silage fermentation quality (pH value of silage, lactic acid (LA), acetic acid (AA), propionic acid (PA), butyric acid (BA) and ammonia nitrogen (NH3-N) content per unit (kg) of silage); and silage nutrients (dry matter (DM), soluble carbohydrate (WSC), crude protein (CP), crude fat (EE), neutral detergent fiber (NDF), acid detergent fiber (ADF), and crude ash (Ash) content per unit (kg) of silage).

[0058] In addition, data from simulated rumen gas production and digestion experiments were used as output variables of the model. Gas production gradually increased over time, with the maximum cumulative total gas production at 24h, 36h, 48h, and 72h being 417, 431.42, 511.2, and 545 mL / g DM, respectively. The accumulation was faster in the early stages and smaller in the later stages. This reflects the general experimental pattern of the effect of diversified silage on rumen gas production. The median dry matter digestibility at multiple time points all exceeded 58%, indicating the rumen's potential for efficient utilization of diversified silage. For the output variables, this invention collected the cumulative total gas production, methane production, dry matter digestibility, neutral detergent fiber digestibility, and acid detergent fiber digestibility at all time points under in vitro rumen experimental conditions. Finally, considering the minimum amount of input data for model training, 14 model output variables were retained as the final prediction targets of the model of this invention (including the digestion results of silage in the rumen of ruminants, including gas production and / or digestibility), including: 24-hour cumulative total gas production (GP_24h), 36-hour cumulative total gas production (GP_36h), 48-hour cumulative total gas production (GP_48h), 72-hour cumulative total gas production (GP_72h), 24-hour cumulative methane production (CH4_24h), 48-hour cumulative methane production (CH4_48h), and 72-hour cumulative methane production. (CH4_72h), 24-hour in vitro rumen dry matter digestibility (IVDMD_24h), 48-hour in vitro rumen dry matter digestibility (IVDMD_48h), 72-hour in vitro rumen dry matter digestibility (IVDMD_72h), 48-hour in vitro rumen neutral detergent fiber digestibility (IVNDFD_48h), 72-hour in vitro rumen neutral detergent fiber digestibility (IVNDFD_72h), 48-hour in vitro rumen acid detergent fiber digestibility (IVADFD_48h), and 72-hour in vitro rumen acid detergent fiber digestibility (IVADFD_72h).

[0059] The data distribution shows that the 27 variables (13 model input variables and 14 model output variables) all have a wide range of distributions (Figure 2). Sufficient variable features and a wide range of values ​​provide ample learning information for the machine learning model. In summary, the diverse experimental data reveal the properties of fermented silage, and the broad and varied variable distribution enhances the model's generalization and practicality during the learning process. The relationship between silage fermentation properties and rumen gas production and digestion is intricate. Traditional statistical analysis cannot unravel this complex relationship; utilizing interpretable machine learning models is an effective approach to unlocking the key to understanding it.

[0060] 1.2 Data interpolation To construct a high-precision optimal model, this invention rigorously selects high-quality experimental data, discards data that does not meet the data imputation quantity requirements, and provides a perfect data frame for each output variable. The MissForest algorithm has been proven to outperform traditional data imputation methods such as mean imputation, ensuring high-quality data imputation without altering the correlation between variables, provided the conditions are met. This approach can improve the upper limit of the model's prediction accuracy.

[0061] To ensure high-quality data imputation, a maximum of three (<30%) missing values ​​for silage indicators are allowed. The MissForest algorithm is used to fill in the missing values. The maximum number of iterations is set to 100, and all calculations use 1000 regression trees.

[0062] 1.3 Model Training and Evaluation To construct the optimal model, this invention first uses the sklearn package in Python (3.10.16) to standardize the final model input data using the StandardScaler method. Next, the standardized model input data is randomly divided into a training set (80%) and a test set (20%). Fourteen prediction models for 14 output variables are constructed using five decision tree-based modeling algorithms (XGBoost, CatBoost, GBR, LightGBM, RF). A grid search method is then used to find the optimal hyperparameter combination, and five-fold cross-validation is used to ensure the model is not overfitted. Finally, the optimal hyperparameters are determined for each modeling algorithm. The coefficient of determination (R²) is used... 2 The model's predictive performance is evaluated using the root mean square error (RMSE) and the root mean square error (RMSE). Figure 1 This is a schematic diagram of the model construction process.

[0063] During the modeling process, this invention performs hyperparameter tuning for each of the five tree models, sacrificing time to obtain the optimal parameter combination for each model, ensuring that the model is both accurate and generalizable. Through grid search, the optimal hyperparameters of the best model were determined (Table 2). The results show that among the 14 optimal models established, the total gas production prediction model and the methane production prediction model have the highest R-values. 2 All could be higher than 0.83, with the highest reaching 0.96; the R-value of the digestibility model 2 All values ​​were above 0.79 except for the IVMMD_24h model (Table 1). This indicates that the above strategies effectively helped the model locate the optimal parameters and the machine learning model achieved high accuracy. This provides a reliable guarantee for exploring the complex relationship between silage fermentation and rumen gas production and digestion from a model perspective.

[0064] Table 1. Performance comparison of five optimal machine learning models on the test set.

[0065] Note: The bold numbers in the table represent the performance metrics of the optimal model.

[0066] Table 2. Optimal Hyperparameters of the Optimal Model

[0067] There are two main metrics for evaluating the performance of a model that solves regression problems: R0 2 and RMSE. 2 The higher the value, the smaller the RMSE, indicating better model performance. By comparing the performance metrics of the optimal prediction models built using five machine learning algorithms for 14 output variables on the test set, the results show that the CatBoost algorithm is best suited for building prediction models for total gas production and methane production (Table 1). Only the CH4_24h model (R... 2 The score (0.95, RMSE 7.85) is slightly lower than the model built using RF (R 2 (The RMSE is 6.33 and the SEM is 0.96). Nevertheless, the performance of the CatBoost model is comparable to that of the RF model, and it can also be used as a production model.

[0068] Regarding digestibility prediction, XGBoost, CatBoost, GBR, and RF all performed best in predicting at least one output variable. Nevertheless, the CatBoost model still performed comparably to the other algorithms. Overall, all five algorithms could construct high-performance models on the experimental data, with CatBoost being the best. Therefore, the CatBoost model was chosen for the final model and feature analysis to evaluate the relationship between silage fermentation and rumen gas production and digestibility.

[0069] The regression plot effectively illustrates the details of the model's performance (Figure 3). Blue and purple represent the model's predictions for the training and test sets, respectively. Both the blue and purple diagonal lines (function y=x) indicate that the predicted values ​​are exactly the same as the true values. The blue and purple shaded areas represent 95% confidence intervals, indicating that the value is more likely to lie within the blue or purple area. The smaller the area of ​​the confidence interval, the lower the level of uncertainty around the data value. As shown in the figure, the CatBoost model performs almost perfectly on the training set because the blue points are almost always on or near the diagonal line (function y=x), resulting in a high R². 2 The R-value is close to 1. However, its performance on the test set is consistently lower than on the training set, which is due to the inherent nature of model training. Among all prediction models, the confidence intervals for the total gas production prediction model and CH4_24h are extremely small, which is one of the reasons why the CatBoost algorithm performs best on gas production data. Nevertheless, a small number of outliers still deviate from the sloping line and confidence interval. This may be because the input features of some data differ from other data, leading to insufficient learning by the machine learning model during training, causing some points on the prediction graph to deviate from the bisector. However, the R-values ​​of the CH4_48h and CH4_72h models, which have insufficient training data, are... 2 All of them are higher than 0.84, while the R of the IVMMD_24h model is... 2 Low (R) 2 =0.58). The regression plot explains this phenomenon: the data ranges for CH4_48h and CH4_72h are small and the data distribution is relatively even, while the vast majority of data points for IVMMD_24h are concentrated in the small range of 450-700 g / kg DM, with very few points distributed in other larger ranges. This leads to insufficient generalization of the model, resulting in performance degradation. In the future, the accumulation of diverse experimental data can help solve this problem.

[0070] 2. Feature Analysis and Model Interpretation To explain how each silage feature influences model predictions, this invention uses feature importance calculation and SHAP analysis to determine their respective impacts. The tree model determines feature importance by recording average feature information acquired throughout training. SHAP values ​​illustrate the relationship between input features and input variables by calculating the marginal impact of each feature on the projected output. Both strategies address the limitations of black-box machine learning models. Furthermore, this invention uses a one-dimensional partial dependency graph (PDP) to explore the average impact of each feature on the machine learning model output across all samples.

[0071] 2.1 Silage drivers influencing rumen gas production, methane generation, and digestion To explore how fermented silage affects the model, feature importance grading (Figure 4) and SHAP analysis were performed. Figure 5 These methods are used to evaluate the importance and impact of input features. Building upon the optimal CatBoost model, by exploring the patterns of change in model output, the driving factors influencing total gas production, methane production, and the three digestibility rates are identified, further enhancing the practical value of machine learning models. Results show that similar results were obtained through two different feature analysis methods, making it more reliable to understand the complex relationships between input and output variables from the model. Tree models recursively split the data, selecting features with the largest information gain or the greatest reduction in variance for node splitting; feature importance is comprehensively evaluated by its contribution to the split (e.g., frequency of use, purity improvement). SHAP, based on the Shapley value in game theory, quantifies the marginal contribution of each feature to the model's predictions, providing a consistent and interpretable measure of feature importance by analyzing the joint effects of all feature subsets. Both provide a compass for explaining how "black box" models operate.

[0072] 2.1.1 Silage drivers affecting total rumen gas production For the total gas production model, the eigenvalue importance of the GP_36h, GP_48h, and GP_72h models all showed that the chemical composition of silage was higher than that of silage fermentation quality, while the GP_24h model showed the opposite (silage fermentation quality was higher than that of silage chemical composition). This indicates that in the early stage of rumen fermentation (within 24 hours), the quality of silage fermentation has a greater impact on gas production. For example, SHAP analysis showed that higher butyric acid and acetic acid content had a negative impact on the model, i.e., reduced gas production. This is related to the fact that these two substances are "immediate substrates" for silage residues. Acetic acid and butyric acid may slow down the generation efficiency of gases such as CH4 and CO2 by altering the rumen microenvironment (such as osmotic pressure and pH gradient) or by directly inhibiting the activity of key gas-producing bacteria such as methanogens and cellulose-degrading bacteria. In the middle and late stages, as the substrate shifts to structural fibers and the function of the microbial community differentiates, the role of organic acids degenerates from "dominant regulation" to "auxiliary metabolism," and their importance is surpassed by long-term limiting factors such as fiber chemical composition, nitrogen sources, and minerals.

[0073] In the GP_36h, GP_48h, and GP_72h models, the chemical composition of silage is more important, with neutral detergent fiber, acid detergent fiber, dry matter, crude fat, and crude ash consistently ranking high in importance. During the middle and late stages of rumen fermentation (36-72h), these chemical components differentially affect gas production through positive and negative effects. Higher levels of neutral and acid detergent fiber result in lower gas production (a negative effect), which is closely related to the structural characteristics of fibrous materials. High levels of neutral or acid detergent fiber are usually accompanied by an increase in lignin and other anti-degradation components, leading to a denser fiber structure, reduced fermentability, and increased energy consumption by rumen microorganisms for decomposition, thus decreasing the actual effective carbon source release efficiency. Furthermore, excessive fiber may physically hinder contact between microorganisms and other fermentable substrates (such as proteins and fats), indirectly inhibiting gas production efficiency. Dry matter and crude fat have a positive effect. Dry matter reflects the total amount of fermentable substrate, and its increase provides microorganisms with more metabolic raw materials. Crude fat, as a high-energy substrate, supplements energy and promotes microbial activity. The negative effect of crude ash may be due to rumen osmotic pressure imbalance caused by excessive minerals or metal ion toxicity, which inhibits microbial enzyme activity.

[0074] Furthermore, ammonia nitrogen has also been found to have a significant impact on rumen gas production in the mid-to-late stages. As a key nitrogen source for microbial protein synthesis, ammonia nitrogen can supplement the needs of microorganisms during their rapid growth phase, promoting the activity and metabolic efficiency of cellulose-decomposing bacteria, methanogens, and other microbial communities. In summary, the importance of these indicators reflects the time dependence of rumen microorganisms on the deep degradation of complex substrates, as well as the synergistic effect of silage fermentation quality (such as ammonia nitrogen) and chemical components in the later stages of fermentation.

[0075] 2.1.2 Silage driving factors affecting rumen methane production CH4 is a significant component of total rumen gas production. Characteristic importance grading shows that the overall importance of silage fermentation quality is higher in the early stage of rumen fermentation (24h), while the overall importance of silage chemical components is higher in the middle and late stages (48h, 72h). This is consistent with the analysis results of total rumen gas production, but there are significant differences in the detailed top importance indicators. The CH4_24h model shows that acetic acid, lactic acid, dry matter, propionic acid, and crude ash have the highest characteristic importance (top five). In the early stage of rumen fermentation, silage fermentation quality indicators (acetic acid, lactic acid, pH) significantly affect CH4 production: Acetic acid, as the main carbon source for methanogens, inhibits methanogen activity by lowering rumen pH (low pH has a negative impact) at high concentrations, or may lead to substrate diversion due to preferential utilization by the host; lactic acid accumulation exacerbates the acidic environment, further inhibiting the coenzyme activity of methanogens and competing for hydrogen substrates, resulting in a decrease in CH4 production. The low positive effects of dry matter and crude ash reflect that soluble substrates (rather than high fiber or high minerals) are more conducive to methanogen utilization in the early stages. Excessive dry matter or crude ash may limit early methane generation due to inefficient substrate degradation or osmotic imbalance.

[0076] As fermentation progresses, the importance of chemical components increases. CH 4_ In the 48h model, high dry matter content had a negative impact, possibly due to slow degradation and insufficient release of effective carbon sources caused by high fiber content. Lactic acid changed from inhibition in the early stage to moderate promotion in the middle stage, indicating that after the rumen pH tended to stabilize, lactic acid, as an intermediate metabolite, could indirectly support the activity of methanogens. 4_ In the 72h model, high values ​​of soluble carbohydrates had a negative impact, suggesting that insufficiently degraded soluble carbohydrates may be accompanied by antinutritional factors, inhibiting the utilization of methanogenic precursors (such as acetic acid and H2) by microorganisms. High values ​​of acetic acid and lactic acid had a positive impact, while butyric acid and propionic acid had a negative impact, reflecting the preference of methanogens for specific volatile fatty acids (mainly acetic acid) in the later stages. Propionic acid metabolism consumes H2, and butyric acid inhibits bacterial activity, leading to CH4 generation differentiation. Overall, CH4 production in the middle and late stages is jointly regulated by "substrate degradation efficiency" (dry matter, soluble carbohydrates) and "volatile fatty acid metabolic balance" (lactic acid to acetic acid ratio), reflecting the functional transition of rumen microorganisms from the adaptation phase to the deep degradation phase. Due to insufficient training data for the CH4_48h and CH4_72h models, the feature analysis is only a simple exploration, and further validation with more experimental data is needed in the future.

[0077] 2.1.3 Silage drivers affecting rumen digestibility Digestibility reflects the degree to which feed nutrients are utilized and absorbed by animals, and is related to economic benefits that farmers care about, such as milk production in dairy cows and weight gain in beef cattle. This invention also found that in the early stage of rumen fermentation (24h), the overall importance of silage fermentation quality is higher than that of chemical components; while in the middle and late stages of fermentation (48h, 72h), the overall importance of chemical components is higher than that of silage fermentation quality. This phenomenon is consistent with the results of rumen gas production and methane production model analysis, suggesting the significant impact of components such as organic acids on the rumen after rumen intake in ruminants. Previous studies have overemphasized the impact of nutrient composition on the rumen, neglecting the importance of these components; one benefit of machine learning is now evident.

[0078] For the early digestion model (IVDMD_24h), the top five most important indicators were acetic acid, lactic acid, dry matter, neutral detergent fiber, and crude ash. SHAP analysis showed that higher levels of acetic acid, lactic acid, and crude ash negatively impacted the prediction of dry matter digestibility. This may be because acetic acid and lactic acid produced during silage fermentation inhibit cellulase activity by lowering rumen pH, and excessive crude ash leads to mineral interference with microbial metabolism, both of which negatively affect early digestibility. The effects of dry matter, neutral detergent fiber, and acid detergent fiber on dry matter digestibility were both positive and negative, making it impossible to draw a definitive conclusion.

[0079] For the intermediate-term digestion model (IVDMD_48h), the top five most important indicators were crude protein, lactic acid, crude fat, acid detergent fiber, and butyric acid. SHAP analysis showed that higher crude protein values ​​had a positive impact on dry matter digestibility prediction; higher lactic acid values ​​had a positive impact on dry matter digestibility prediction; higher crude fat values ​​had a negative impact on dry matter digestibility prediction; and higher acid detergent fiber values ​​had a negative impact on dry matter digestibility prediction. This may be because the ammonia nitrogen provided by crude protein degradation promotes microbial enzyme secretion, and the decrease in lactic acid concentration maintains a suitable pH environment, both of which positively enhance digestibility; crude fat coats feed particles, forming a physical barrier, while acid detergent fiber, due to its high lignin binding degree, increases the difficulty of degradation, leading to a negative effect.

[0080] For the late-stage digestibility model (IVDMD_72h), the top five most important indicators were crude ash, dry matter, neutral detergent fiber, acetic acid, and crude protein. SHAP analysis showed that a higher dry matter value positively impacted dry matter digestibility prediction. The effects of crude protein, crude ash, neutral detergent fiber, and acetic acid were mixed, making it impossible to draw definitive conclusions. Dry matter content determines the total amount of remaining fermentable substrate, and its high value positively influences digestibility through its continued degradation potential. The effects of indicators such as neutral detergent fiber and crude ash exhibit a bidirectional effect due to substrate structural heterogeneity (e.g., fiber density, mineral balance), reflecting that microbial degradation of complex substrates has entered a plateau phase, limited by enzyme efficiency and environmental homeostasis.

[0081] For the IVNDFD_48h model, the top five most important indicators were dry matter, ammonia nitrogen, propionic acid, butyric acid, and acid detergent fiber. SHAP analysis showed that higher values ​​for dry matter, ammonia nitrogen, propionic acid, butyric acid, and acid detergent fiber all negatively impacted the prediction of neutral detergent fiber digestibility. High dry matter values ​​may correspond to high lignin binding or dense structure in neutral detergent fibers, increasing resistance to microbial degradation; excessive ammonia nitrogen may inhibit fiber-decomposing bacteria (such as...). Ruminococcus Activity; high concentrations of propionic acid and butyric acid inhibit cellulase secretion through feedback regulation; acid detergent fiber and neutral detergent fiber are highly correlated, and their increase directly reflects the increase in the proportion of recalcitrant components (such as lignin) in the fiber, which together lead to a decrease in 48h digestibility.

[0082] For the IVNDFD_72h model, the top five most important features were butyric acid, dry matter, lactic acid, neutral detergent fiber, and pH. SHAP analysis showed that higher butyric acid, lactic acid, and pH values ​​negatively impacted the prediction of neutral detergent fiber digestibility, while higher dry matter and neutral detergent fiber values ​​positively impacted it. High concentrations of butyric acid and lactic acid may inhibit the activity of fiber-degrading bacteria by lowering pH, or promote competitive microbial communities (such as those utilizing butyric acid). Butyrivibrio (1) It can crowd out ecological niches; a high dry matter value indicates that the total amount of fermentable substrate is sufficient, and a moderate increase in neutral detergent fiber may reflect that the fiber structure is loose (easily degradable). During long-term fermentation, microorganisms improve digestibility through continuous enzymatic hydrolysis.

[0083] For the IVADFD_48h model, the top five most important indicators were acetic acid, propionic acid, soluble carbohydrates, acid detergent fiber, and ammonia nitrogen. SHAP analysis showed that higher values ​​of acetic acid, propionic acid, acid detergent fiber, and ammonia nitrogen negatively impacted the prediction of acid detergent fiber digestibility, while higher values ​​of soluble carbohydrates positively impacted it. High concentrations of acetic acid and propionic acid may slow down acid detergent fiber degradation by inhibiting cellulase activity (e.g., conformational changes in cellulase). Acid detergent fiber itself contains a high proportion of lignin, and its increase directly increases the structural barrier. Excessive ammonia nitrogen causes nitrogen toxicity, inhibiting microbial proliferation. High values ​​of soluble carbohydrates promote rapid microbial proliferation and secretion of cellulase-degrading enzymes (e.g., xylanase), indirectly improving 48h digestibility.

[0084] For the IVADFD_72h model, the top five most important indicators were acid detergent fiber, acetic acid, dry matter, neutral detergent fiber, and ammonia nitrogen. SHAP analysis showed that a higher dry matter value positively impacted the prediction of acid detergent fiber digestibility. The effects of acid detergent fiber, neutral detergent fiber, acetic acid, and ammonia nitrogen on acid detergent fiber digestibility were mixed, but most low values ​​had a positive impact on model prediction, while a small number of extremely high values ​​had a negative impact. High dry matter values ​​provided sufficient substrate to support long-term degradation. Extremely low acid detergent fiber / neutral detergent fiber ratios may correspond to insufficient fiber content, resulting in a lack of sustained targets for microbial action, while extremely high values ​​resulted in an "anti-degradation matrix" due to excessive lignin cross-linking. Extremely low acetic acid and ammonia nitrogen values ​​may lead to carbon-nitrogen imbalance and inhibit enzyme secretion, while extremely high values ​​reduced digestibility due to feedback inhibition or toxic effects, reflecting the nonlinear relationship between fiber digestibility and the "substrate concentration-degradation threshold".

[0085] In summary, considering the differences in objectives, total gas production is driven by the cumulative effect of broad-spectrum substrate utilization, CH4 production is constrained by the specific substrate preference of methanogens, and digestibility is determined by the dynamic balance between the fiber structural barrier and enzyme activity. Fermentation quality indicators dominate the immediate effects in the early stages, while chemical composition indicators highlight long-term degradation potential in the later stages. These two factors, through the synergy and balancing of nitrogen sources (ammonia nitrogen), energy (crude fat), and structural substrates (fiber), form a differentiated regulatory mechanism for different predicted objectives. In conclusion, this invention, based on an interpretable model, learned the patterns of experimental big data and explained from a model perspective how silage fermentation affects rumen gas production, methane generation, and digestion.

[0086] 2.2 Effects of silage driving factors on rumen gas production, methane yield, and digestibility The partial dependency plot shows the average marginal effect of one or two feature variables on the model's prediction results, indicating how changes in the values ​​of specific feature variables affect the model's predicted output, thus providing a dynamic perspective for this invention's study of the effects of silage fermentation properties on rumen fermentation gas production and digestion. Since this invention generates more than 150 input-output variable PDP analysis results, only a portion of the results are presented and described (Figure 6).

[0087] 2.2.1 Effect of pH value Partial dependence analysis showed a negative correlation with low pH. When pH was between 3.79 and 4.25, it showed a strong negative correlation with the cumulative total gas production at multiple time points; increasing pH drastically reduced gas production. This phenomenon was also observed in the prediction of rumen digestibility (IVDMD_72h, IVADFD_72h, IVNDFD_72h, etc.). This may be because the intake of low-pH silage alters the rumen microenvironment (e.g., osmotic pressure, pH gradient), inhibiting the activity of key gas-producing bacteria such as methanogens and cellulose-degrading bacteria, as well as digestive enzymes. The concentration of organic acids in silage can significantly affect the rumen within a certain range. For example, a high lactic acid content can increase gas production, methane production, and digestibility. This is reasonable, as high lactic acid provides energy for rumen microorganisms, producing CO2 and H2 through metabolism, which directly produce gas, provide raw materials for methanogens, and enhance the activity of related bacteria, increasing gas production and CH4 production. Lactic acid metabolites nourish rumen microorganisms, promote community synergy, enhance the digestion and absorption of various nutrients in silage, and further improve digestibility. This invention found that low levels of ammonia nitrogen have a drastic impact on total rumen gas production over 24 hours, while in the GP_72h model, it shows a stable negative impact on total rumen gas production over 72 hours. In the early stages of rumen fermentation, at low ammonia nitrogen levels, microorganisms rapidly adjust their metabolism due to changes in nitrogen source. Competition for resources among microbial species and enzyme activity affected by nitrogen concentration fluctuations cause gas production to initially increase suddenly due to the proliferation of some bacteria and accelerated enzymatic reactions, followed by a sudden decrease due to resource competition and changes in enzyme activity. However, towards the end of fermentation, the microbial community has stabilized. At low ammonia nitrogen levels, an increase in ammonia nitrogen concentration disrupts the ecological balance of the community, leading to the proliferation of other microorganisms that crowd out the living space of key gas-producing bacteria, thus affecting gas production. This aligns with general empirical patterns.

[0088] 2.2.2 Effects of Nutritional Components The nutrient composition of silage also has a significant impact on rumen gas production. In the 60-120 g / kg DM range, crude protein has a negative impact on total rumen gas production over 24 hours. This may be because microorganisms preferentially decompose protein, consuming energy and nitrogen sources, reducing carbohydrate fermentation and gas production. Furthermore, protein fermentation raises the pH, inhibiting the activity of gas-producing bacteria. However, it has a positive impact on CH4 production over 24 hours because protein decomposition products provide substrates such as methyl donors for methanogens, while simultaneously promoting the proliferation of methanogens and their symbiotic communities, thus increasing CH4 synthesis efficiency. The difference between these two effects stems from the metabolic allocation of nitrogen sources by microorganisms (protein vs. carbohydrates) and the functional differentiation of the microbial community (inhibition of gas-producing bacteria vs. promotion of methanogens). This explanation was validated in the analysis of soluble carbohydrates. When the soluble carbohydrate content is greater than 20 g / kg DM, it has a positive impact on total rumen gas production over 24 hours. When the soluble carbohydrate content is greater than 20 g / kg DM, this inhibition is compensated by providing sufficient fermentable carbon sources. This positive impact stems from the preferential metabolic selection of microorganisms for high-concentration carbohydrates. Therefore, when carbon sources (soluble carbohydrates) are sufficient, the bacterial community tends to decompose carbohydrates to quickly produce acid and gas, while the nitrogen source effect of crude protein focuses more on supporting methanogens rather than overall gas production in the early stages, eventually forming a metabolic division of labor of "soluble carbohydrates driving gas production and crude protein regulating CH4".

[0089] 2.2.3 Influence of Fibers Numerous studies have shown that dietary fiber has a significant impact on the rumen of ruminants. Ruminants (such as cattle and sheep) rely on fiber to maintain the physical structure of the rumen and rumination behavior—fiber stimulates chewing and saliva secretion, promoting the mixing and grinding of rumen contents and creating conditions for subsequent digestive enzyme action. This process is closely related to the fiber's own structure (acid detergent fiber / neutral detergent fiber) and its influence on microbial degradation, jointly determining the dynamic changes in rumen gas production and digestion. The results of this invention support this conclusion and quantify the impact. In all rumen gas production model analyses, both acid detergent fiber and neutral detergent fiber had a negative impact on the model output; however, acid detergent fiber showed a positive impact on the CH4 production model. Acid detergent fiber and neutral detergent fiber, due to their content of lignin and other recalcitrant components, physically hinder fiber decomposition and increase metabolic energy consumption, leading to a decrease in total rumen gas production as their content increases. The H2 and CO2 produced by cellulose degradation in acid detergent fiber are key substrates for methanogens; their high content promotes CH4 generation by strengthening the "hydrogen production-hydrogen consumption" symbiotic relationship. The difference in the impact on total gas production and CH4 production is essentially due to the different effects of structural fibers on the overall metabolic efficiency of microorganisms and the CH4 generation pathway. For rumen digestion, both acid detergent fiber and neutral detergent fiber showed a negative impact on the 24-hour and 48-hour predictions of rumen dry matter digestibility. However, they showed a positive impact on the 72-hour prediction of rumen dry matter digestibility, especially when the neutral detergent fiber content was greater than 400 g / kg DM. In the early and mid-stages of digestion, lignin in acidic / neutral detergent fiber forms a physical barrier, inhibiting the attachment and enzymatic degradation of fiber-decomposing bacteria. Furthermore, microorganisms preferentially utilize soluble substrates, leading to insufficient degradation of structural fibers and slowing down dry matter digestion. However, in the later stages of digestion (72 hours), with the proliferation of fiber-decomposing bacteria and continuous enzyme secretion, the lignin-encapsulated fiber structure gradually disintegrates. Especially when neutral detergent fiber content exceeds 400 g / kg DM, sufficient structural substrate stimulates long-term metabolic adaptation of microorganisms, promoting dry matter digestion. This difference essentially represents a functional shift in rumen microorganisms from "substrate competition limited by physical barriers in the early stages" to "adaptive degradation dominated by structural fibers in the later stages."

[0090] Therefore, biased graphs provide a dynamic analytical perspective for interpretive machine learning, enabling the intuitive identification of the differential effects of silage fermentation indicators (such as pH, lactic acid, acid detergent fiber, etc.) on rumen gas production, CH4 production, and digestibility at different time scales. This helps extract interpretable biological laws from the output of complex models and provides a data-driven scientific basis for optimizing ruminant feed formulations.

[0091] 3. Comparison between the optimal model and empirical formulas Currently, several output targets can generally be predicted by empirical formulas. Here, this invention predicts the optimal model and the empirical formula (%DMD=88.9-(0.779×%ADF)) (with additional data (not involved in the model building process)), and compares the difference between the predicted and experimental values ​​by calculating the mean absolute error (MAPE) to evaluate whether machine learning is superior to empirical paradigms.

[0092] To evaluate the practical application of the model of this invention, the optimal model corresponding to each output variable was used to compute three additional experimental data points (these experimental data points are independent of the training and test sets). Figure 10 ).

[0093] The results from the three experimental data ( Figure 9 The comparison of the true and predicted values ​​in Figure 7 shows that the average absolute error of all models does not exceed 27%, with the IVMMD_72h model showing an even lower average absolute error of 3.11%, indicating that the model of this invention has high practicality and accuracy. An empirical formula for calculating rumen dry matter digestibility based solely on acid detergent fiber content was also used to predict the output for the same data points, and the results showed that the average absolute error of the empirical formula was 11.21%. Therefore, the IVMMD_72h model of this invention is significantly superior to the empirical formula.

[0094] In summary, this invention innovatively utilizes machine learning to analyze the complex relationship between silage fermentation and rumen gas production, methane production, and digestibility. The optimal model prediction method and product developed in this invention can replace traditional time-consuming and costly rumen experiments. This not only protects experimental animals to a certain extent but also guides farmers and herdsmen to formulate feed in a targeted manner, thereby achieving the vision of precise feeding, cost reduction and efficiency improvement, and environmental protection.

[0095] The present invention has been described in detail above. Those skilled in the art will recognize that the invention can be practiced in a wide range of ways with equivalent parameters, concentrations, and conditions without departing from its spirit and scope, and without requiring unnecessary experiments. While specific embodiments have been provided, it should be understood that further modifications can be made to the invention. In summary, according to the principles of the invention, this application is intended to include any changes, uses, or improvements to the invention, including changes made using conventional techniques known in the art that depart from the scope disclosed herein.

Claims

1. A data processing apparatus, comprising a memory, a processor, and a computer program stored in the memory, characterized in that: The processor executes the computer program to perform the following steps: A1) Model Construction: Based on 13 model input variables and model output variables, a machine learning algorithm was used to construct a model to predict the gas production and digestion of silage in the rumen of ruminants. The 13 model input variables are the pH of the silage, the lactic acid content, acetic acid content, propionic acid content, butyric acid content, ammonia nitrogen content, dry matter content, soluble carbohydrate content, crude protein content, crude fat content, neutral detergent fiber content, acid detergent fiber content, and crude ash content of the silage. The model output variables are the gas production and / or digestibility of the silage in the rumen. A2) Model prediction output: Input the 13 model input variables of the silage to be tested into the model obtained in A1), and output the gas production and digestion results of the silage to be tested in the rumen of ruminants, the gas production and digestion results including gas production and / or digestibility.

2. The data processing apparatus according to claim 1, characterized in that: The gas production includes total gas production and / or methane production; the digestibility includes dry matter digestibility and / or detergent fiber digestibility; the detergent fiber includes neutral detergent fiber and / or acid detergent fiber.

3. The data processing apparatus according to claim 1 or 2, characterized in that: The machine learning algorithm used is CatBoost.

4. Methods for predicting the gas production and digestion of silage in the rumen, including: B1) Model Construction: Based on 13 model input variables and model output variables, a machine learning algorithm was used to construct a model to predict the gas production and digestion of silage in the rumen of ruminants. The 13 model input variables are the pH of the silage, the lactic acid content, acetic acid content, propionic acid content, butyric acid content, ammonia nitrogen content, dry matter content, soluble carbohydrate content, crude protein content, crude fat content, neutral detergent fiber content, acid detergent fiber content, and crude ash content of the silage. The model output variables are the gas production and / or digestibility of the silage in the rumen. B2) Model prediction output: Input the 13 model input variables of the silage to be tested into the model obtained in B1), and output the gas production and digestion results of the silage to be tested in the rumen of ruminants, the gas production and digestion results including gas production and / or digestibility.

5. The method according to claim 4, characterized in that: The gas production includes total gas production and / or methane production; the digestibility includes dry matter digestibility and / or detergent fiber digestibility; the detergent fiber includes neutral detergent fiber and / or acid detergent fiber.

6. The method according to claim 4 or 5, characterized in that: The machine learning algorithm used is CatBoost.

7. A device for predicting the gas production and digestion of silage in the rumen, characterized in that: The device includes the following modules: C1) Model Building Module: Used to construct a model predicting the gas production and digestibility of silage in the rumen of ruminants using machine learning algorithms, based on 13 model input variables and model output variables. The 13 model input variables are the pH of the silage, the lactic acid content, acetic acid content, propionic acid content, butyric acid content, ammonia nitrogen content, dry matter content, soluble carbohydrate content, crude protein content, crude fat content, neutral detergent fiber content, acid detergent fiber content, and crude ash content of the silage; the model output variables are the gas production and / or digestibility of the silage in the rumen. C2) Model prediction output module: used to input the 13 model input variables of the silage to be tested into the model obtained in C1), and output the gas production and digestion results of the silage to be tested in the rumen of ruminants, the gas production and digestion results including gas production and / or digestibility.

8. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, it implements the steps of the method described in any one of claims 4-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the steps of the method described in any one of claims 4-6.

10. Any of the following applications of the data processing apparatus according to any one of claims 1-3 and / or the computer program product according to claim 8 and / or the computer-readable storage medium according to claim 9: C1) Application in the preparation of green feed for ruminants; Application of C2 in the preparation of precision feeding products for ruminants.

Images