Processing preparation technology for production of high-quality cattle and sheep feed

Abstract

The invention discloses a processing preparation technology for high-quality cattle and sheep feed production, and relates to the technical field of livestock feed processing, the specific steps are as follows: S1, raw material pretreatment: screening raw materials and removing impurities; s2, intelligent formula calculation: establishing a multi-objective optimization model; s3, mixing the raw materials: putting the raw materials into a double-shaft paddle mixer according to the optimal proportion; s4, tempering-extrusion granulation: establishing a processing technology parameter self-adaptive regulation and control model; s5, quality on-line monitoring: establishing a quality on-line monitoring and feedback model; s6, cooling and screening: performing countercurrent cooling on the particles to room temperature, and removing broken particles by a classifying screen; and S7, finished product packaging: packaging according to specifications, and warehousing for storage. According to the processing and preparing technology for producing the high-quality cattle and sheep feed, the nutrition precision is remarkably improved, nutrition-cost-digestibility collaborative optimization is achieved through a multi-objective optimization model, the feed nutrition satisfaction degree is improved, and the situation that a single nutrient is excessive or insufficient is avoided.

Description

Technical Field

[0001] This application relates to the field of livestock feed processing technology, and more specifically, to a processing and preparation process for the production of high-quality cattle and sheep feed. Background Technology

[0002] As cattle and sheep farming develops towards large-scale and intensive operations, the demand for "high-quality" feed is becoming increasingly prominent. This not only requires that the nutritional components (crude protein, crude fiber, calcium-to-phosphorus ratio, etc.) meet the physiological needs of cattle and sheep, but also needs to consider digestibility, mycotoxin control, and cost-effectiveness. However, existing cattle and sheep feed preparation technologies suffer from the following core problems:

[0003] The formulation design is crude: Traditional formulations rely on empiricism and adjust the proportion of raw materials only through a single nutritional indicator (such as crude protein content). They do not consider the synergistic / antagonistic effects between nutrients (such as excessive calcium inhibiting phosphorus absorption) and the dynamic needs of cattle and sheep at different growth stages, resulting in unbalanced feed nutrition and low utilization rate.

[0004] Blindness in process parameters: The parameters (temperature, speed, moisture) of key processing links such as extrusion granulation and conditioning rely on manual trial and error to determine, which cannot cope with the process fluctuations caused by raw material moisture fluctuations and batch differences. This can easily lead to uneven particle hardness, insufficient gelatinization or excessive gelatinization, resulting in nutrient loss.

[0005] Lagging quality monitoring: Current testing is mostly done offline (such as laboratory testing for crude protein and mycotoxins), with testing cycles taking 2-4 hours. This makes it impossible to provide real-time feedback and adjust the production process, resulting in substandard feed flowing into subsequent stages and a high defect rate.

[0006] Therefore, a processing and preparation technology for the production of high-quality cattle and sheep feed is proposed to address the aforementioned technical issues. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the prior art, this application provides a processing and preparation process for the production of high-quality cattle and sheep feed to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, this application provides the following technical solution: a processing and preparation process for producing high-quality cattle and sheep feed, comprising the following specific steps:

[0009] S1. Raw material pretreatment: Screen the raw materials and remove impurities, dry them at low temperature to a moisture content of 12-14%, and crush them to a particle size of 1.0-1.5mm. The raw materials include, but are not limited to, corn, soybean meal, alfalfa, wheat bran, and premixed feed.

[0010] S2. Intelligent formulation calculation: Establish a multi-objective optimization model, which includes objective functions aimed at maximizing nutritional satisfaction, minimizing cost, and maximizing digestibility. Solve the optimal raw material ratio by improving the NSGA-II algorithm. The constraints of the multi-objective optimization model include raw material usage, nutritional safety, and physical property constraints.

[0011] S3. Raw material mixing: Add raw materials to a twin-shaft paddle mixer according to the optimal ratio, mix at a speed of 200 r / min for 5 min, and the coefficient of variation of mixing uniformity should be ≤7%;

[0012] S4. Conditioning-Extrusion Granulation: An adaptive control model for processing parameters is established, and the adaptive control model adopts a PID-neural network fusion control algorithm. The conditioning temperature is set and dynamically adjusted to 85-90℃, the extrusion speed to 280-320r / min, and the conditioning moisture content to 14-16% based on the real-time moisture content and particle size of the raw materials.

[0013] S5. Online Quality Monitoring: Establish an online quality monitoring and feedback model. The online monitoring and feedback model collects particle spectral data through a near-infrared spectrometer and combines it with the PLSR model to detect the content of crude protein, crude fiber, and mycotoxins in real time.

[0014] S6. Cooling and screening: The particles are cooled to room temperature by countercurrent cooling, and broken particles are removed by a grading sieve;

[0015] S7. Finished product packaging: Pack according to specifications, store in the warehouse, and the storage conditions are temperature ≤25℃ and relative humidity ≤60%.

[0016] Furthermore, in S2, the objective function for maximizing nutritional satisfaction is used to measure the degree of matching between the key nutrients in the feed (crude protein CP, crude fiber CF, calcium Ca, phosphorus P, vitamin AVA, etc.) and the requirements of cattle and sheep. The key nutrients include, but are not limited to, crude protein CP, crude fiber CF, calcium Ca, phosphorus P, and vitamin AVA. Specifically, the objective function for maximizing nutritional satisfaction is as follows: ,in Key nutrient types; For the first The first of the raw materials The content of various nutrients; For target cattle and sheep (such as fattening beef cattle and lactating dairy cows) Nutrient requirements standards (refer to NY / T815-2021 "Beef Cattle Feeding Standards" and NY / T34-2004 "Dairy Cattle Feeding Standards"); For the first The quality percentage of each type of raw material; The closer the value is to 1, the more balanced the nutrition.

[0017] Furthermore, the objective function for minimizing costs in S2 focuses on raw material procurement costs, specifically as follows: , For the first The unit cost of the raw materials.

[0018] Furthermore, in S2, the objective function for maximizing nutrient digestibility considers the synergistic effect of raw material combinations on digestibility (e.g., combining soybean meal with alfalfa can improve protein digestibility), and a digestibility synergy coefficient is introduced. Specifically: ,in For the first The first of the raw materials Basic digestibility of various nutrients; For the first , The combination of raw materials for the first The digestibility synergy coefficient of each nutrient, with a positive coefficient indicating synergistic enhancement and a negative coefficient indicating antagonism; For the first The weight of each nutrient (set according to the importance of the needs of cattle and sheep, such as the weight of calcium in lactating dairy cows being 0.25).

[0019] Furthermore, the specific constraints of the S2 multi-objective optimization model are as follows:

[0020] Raw material usage constraints: ,in For the first Minimum addition amount of each raw material; For the first The maximum amount of each raw material to be added;

[0021] Nutritional safety constraints: ,in For the first Mycotoxin content in the raw materials For toxicity limits;

[0022] Physical property constraints: ,in For the first The moisture content of the raw materials, This represents the upper limit of moisture content in the mixed raw materials.

[0023] Furthermore, the improved NSGA-II algorithm in S2 introduces an adaptive crossover and mutation operator, the specific steps of which are as follows:

[0024] B1. Initialize the population: Generate A raw material proportioning scheme that meets the constraints;

[0025] B2. Fast Non-Dominated Ranking: Calculate the non-dominated level (Pareto level) and crowding degree of each individual to select high-quality individuals;

[0026] B3. Adaptive genetic operations:

[0027] Cross operator : ,in This represents the current iteration number. The maximum number of iterations, and In the early stages of iteration, a high crossover probability of 0.8 ensures diversity, while in the later stages, a low crossover probability of 0.6 promotes convergence.

[0028] Mutation Operator : The high mutation probability of 0.2 in the later stages of iteration avoids local optima;

[0029] B4. Elite Retention Strategy: Merge parent and child generations, and select the next generation. A high-quality individual enters the next generation;

[0030] B5. Termination condition: Iteration to... Output the optimal solution set for non-dominant levels, and select the final allocation scheme according to user needs (such as prioritizing cost reduction or improving digestibility).

[0031] Furthermore, the key parameters controlled by the PID-neural network fusion algorithm in S4 include: tempering temperature. (Affecting starch gelatinization degree), extruder speed (Affecting particle density), conditioning moisture (Affects particle hardness); The target for process parameter control is: starch gelatinization degree. (Improving energy digestibility); Particle hardness (To avoid breakage during transportation and ensure palatability); Pellet formation rate ;

[0032] The PID-neural network fusion algorithm structure consists of a prediction layer (BP neural network prediction model) and a control layer (PID closed-loop control model). The specific strategy is as follows:

[0033] The BP neural network prediction model is as follows:

[0034] Input layer: 3 neurons, corresponding to the real-time moisture content of the raw materials. Raw material particle size Target degree of gelatinization ;

[0035] Hidden layer: 10 neurons, activation function is Sigmoid function;

[0036] Output layer: 3 neurons, corresponding to the theoretical values ​​of the optimal process parameters. , These are the optimal conditioning temperature, optimal extruder speed, and optimal conditioning moisture content, respectively.

[0037] Training process: Gradient descent is used to minimize the prediction error. ,and ,in These are the actual process parameters. For the predicted value, iterate to ;

[0038] The PID closed-loop control model is specifically based on the output of the BP neural network prediction model. The PID setpoint is used to collect the actual values ​​of process parameters in real time. The control quantity is calculated using the positional PID formula, specifically: ,in ; These are the PID parameters (determined using the Ziegler-Nichols tuning method). For control quantities (such as heating power adjustment quantities);

[0039] Dynamic correction: The actual values ​​of process parameters and quality indicators (gelatinization degree, hardness) are collected every 5 seconds and fed back to the BP neural network prediction model to update the prediction model parameters.

[0040] Furthermore, the online monitoring and feedback model in S5 monitors the key quality indicators of the finished feed product in real time, and the key quality indicators include crude protein. Coarse fiber Mycotoxins When the online monitoring and feedback model collects particle spectral data using a near-infrared spectrometer, a diffuse reflectance near-infrared spectrometer is installed at the outlet of the extrusion pellet mill to collect spectral data of the feed particles every 2 seconds. (dimension) (corresponding to absorbance at 1500 wavelength points), and then combined with the PLSR model to detect crude protein, crude fiber, and mycotoxin content in real time;

[0041] The PLSR model is used to establish a linear relationship between spectral data and quality indicators. The specific strategy is as follows:

[0042] C1. Sample set construction: Collect 1000 feed samples and measure their spectral data. Quality indicators compared with laboratory testing ;

[0043] C2. Data Preprocessing: [This section appears to be incomplete and requires further context.] Multivariate scattering correction, i.e., MSC and first derivative processing, is performed to eliminate interference caused by particle size and surface scattering.

[0044] C3.PLSR Modeling: Principal Component Extraction: Extraction via PLSR Each principal component, Dimensional reduction Establish a regression model : ,in The regression coefficient matrix , The residual matrix is ​​used for model validation, which involves cross-validation to obtain the coefficients of determination. Root mean square error This meets the monitoring accuracy requirements.

[0045] Furthermore, in S5, when the quality indicator deviates from the target range during online monitoring, a feedback model is triggered. The specific strategy of the feedback model is as follows:

[0046] like Feedback to the multi-objective optimization model in S2 increases the proportion of protein raw materials such as soybean meal and fish meal, specifically by adjusting the amount. ,in This represents the actual crude protein value. This represents the target value for crude protein.

[0047] like The adaptive control model for processing parameters is fed back to S4 to improve the setpoint for tempering temperature. ;

[0048] like Immediately trigger an alarm, halt production, and investigate raw material contamination issues. This represents the actual mycotoxin value.

[0049] Furthermore, in S4, the extrusion granulation adopts a ring die extrusion granulator with a ring die diameter of 6-10mm.

[0050] The technical effects and advantages of this application are as follows:

[0051] Compared with existing technologies, this processing and preparation technology for the production of high-quality cattle and sheep feed significantly improves nutritional precision: it achieves synergistic optimization of nutrition, cost and digestibility through a multi-objective optimization model, improves the nutritional satisfaction of feed, and avoids excessive or insufficient amounts of a single nutrient.

[0052] By introducing a raw material synergistic digestibility coefficient, the crude protein digestibility is effectively improved. A PID-neural network fusion algorithm is employed to achieve dynamic control of the core "conditioning-extrusion granulation" process. Specific advantages include: more sensitive parameter adjustment, more stable finished product quality, lower operational dependence, and a shorter detection cycle for indicators such as crude protein, crude fiber, and mycotoxins.

[0053] By reducing raw material procurement costs through a multi-objective optimization model, combined with improved finished product qualification rates and reduced rework of broken grains, the overall feed production cost is reduced; the scenario adaptation is more flexible: only the nutritional requirement parameters in the multi-objective optimization model need to be adjusted to quickly adapt to cattle and sheep at different stages such as fattening, lactation, and pregnancy, without the need to modify the production line structure. Attached Figure Description

[0054] Figure 1 This is a flowchart illustrating the method of this application;

[0055] Figure 2 This is a schematic diagram of the method flow of the improved NSGA-II algorithm of this application;

[0056] Figure 3 This is a schematic diagram of the strategy flow of the PLSR model in this application. Detailed Implementation

[0057] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0058] Example: As attached Figures 1-3 The processing and preparation technology shown is for the production of high-quality cattle and sheep feed, and the specific steps are as follows:

[0059] S1. Raw material pretreatment: Screen the raw materials and remove impurities, dry them at low temperature to a moisture content of 12-14%, and crush them to a particle size of 1.0-1.5mm. The raw materials include, but are not limited to, corn, soybean meal, alfalfa, wheat bran, and premixed feed.

[0060] S2. Intelligent formulation calculation: Establish a multi-objective optimization model, which includes objective functions aimed at maximizing nutritional satisfaction, minimizing cost, and maximizing digestibility. Solve the optimal raw material ratio by improving the NSGA-II algorithm. The constraints of the multi-objective optimization model include raw material usage, nutritional safety, and physical property constraints.

[0061] In S2, the objective function of maximizing nutritional satisfaction is used to measure the degree of matching between the key nutrients in the feed (crude protein CP, crude fiber CF, calcium Ca, phosphorus P, vitamin AVA, etc.) and the requirements of cattle and sheep. The key nutrients include, but are not limited to, crude protein CP, crude fiber CF, calcium Ca, phosphorus P, and vitamin AVA. The specific objective function for maximizing nutritional satisfaction is as follows: ,in Key nutrient types; For the first The first of the raw materials The content of various nutrients; For target cattle and sheep (such as fattening beef cattle and lactating dairy cows) Nutrient requirements standards (refer to NY / T815-2021 "Beef Cattle Feeding Standards" and NY / T34-2004 "Dairy Cattle Feeding Standards"); For the first The quality percentage of each type of raw material; The closer the value is to 1, the more balanced the nutrition.

[0062] In S2, the objective function for minimizing costs focuses on raw material procurement costs, specifically: , For the first The unit cost of the raw materials.

[0063] In S2, the objective function of maximizing nutrient digestibility considers the synergistic effect of raw material combinations on digestibility (e.g., combining soybean meal with alfalfa can improve protein digestibility), and introduces a digestibility synergy coefficient. Specifically: ,in For the first The first of the raw materials Basic digestibility of various nutrients; For the first , The combination of raw materials for the first The digestibility synergy coefficient of each nutrient, with a positive coefficient indicating synergistic enhancement and a negative coefficient indicating antagonism; For the first The weight of each nutrient (set according to the importance of the needs of cattle and sheep, such as the weight of calcium in lactating dairy cows being 0.25).

[0064] The specific constraints of the S2 multi-objective optimization model are as follows:

[0065] Raw material usage constraints: ,in For the first Minimum addition amount of each raw material; For the first The maximum amount of each raw material to be added;

[0066] Nutritional safety constraints: ,in For the first Mycotoxin content in the raw materials For toxicity limits;

[0067] Physical property constraints: ,in For the first The moisture content of the raw materials, This represents the upper limit of moisture content in the mixed raw materials.

[0068] The improved NSGA-II algorithm in S2 introduces an adaptive crossover and mutation operator. The specific steps are as follows:

[0069] B1. Initialize the population: Generate A raw material proportioning scheme that meets the constraints;

[0070] B2. Fast Non-Dominated Ranking: Calculate the non-dominated level (Pareto level) and crowding degree of each individual to select high-quality individuals;

[0071] B3. Adaptive genetic operations:

[0072] Cross operator : ,in This represents the current iteration number. The maximum number of iterations, and In the early stages of iteration, a high crossover probability of 0.8 ensures diversity, while in the later stages, a low crossover probability of 0.6 promotes convergence.

[0073] Mutation Operator : The high mutation probability of 0.2 in the later stages of iteration avoids local optima;

[0074] B4. Elite Retention Strategy: Merge parent and child generations, and select the next generation. A high-quality individual enters the next generation;

[0075] B5. Termination condition: Iteration to... Output the optimal solution set for non-dominant levels, and select the final allocation scheme according to user needs (such as prioritizing cost reduction or improving digestibility).

[0076] S3. Raw material mixing: Add raw materials to a twin-shaft paddle mixer according to the optimal ratio, mix at a speed of 200 r / min for 5 min, and the coefficient of variation of mixing uniformity should be ≤7%;

[0077] S4. Conditioning-Extrusion Granulation: An adaptive control model for processing parameters is established, and the adaptive control model adopts a PID-neural network fusion control algorithm. The conditioning temperature is set and dynamically adjusted to 85-90℃, the extrusion speed to 280-320r / min, and the conditioning moisture content to 14-16% based on the real-time moisture content and particle size of the raw materials.

[0078] Key parameters controlled by the PID-neural network fusion algorithm in S4 include: tempering temperature. (Affecting starch gelatinization degree), extruder speed (Affecting particle density), conditioning moisture (Affects particle hardness); The target for process parameter control is: starch gelatinization degree. (Improving energy digestibility); Particle hardness (To avoid breakage during transportation and ensure palatability); Pellet formation rate ;

[0079] The PID-neural network fusion algorithm structure consists of a prediction layer (BP neural network prediction model) and a control layer (PID closed-loop control model). The specific strategy is as follows:

[0080] The BP neural network prediction model is as follows:

[0081] Input layer: 3 neurons, corresponding to the real-time moisture content of the raw materials. Raw material particle size Target degree of gelatinization ;

[0082] Hidden layer: 10 neurons, activation function is Sigmoid function;

[0083] Output layer: 3 neurons, corresponding to the theoretical values ​​of the optimal process parameters. , These are the optimal conditioning temperature, optimal extruder speed, and optimal conditioning moisture content, respectively.

[0084] Training process: Gradient descent is used to minimize the prediction error. ,and ,in These are the actual process parameters. For the predicted value, iterate to ;

[0085] The PID closed-loop control model is specifically based on the output of the BP neural network prediction model. The PID setpoint is used to collect the actual values ​​of process parameters in real time. The control quantity is calculated using the positional PID formula, specifically: ,in ; These are the PID parameters (determined using the Ziegler-Nichols tuning method). For control quantities (such as heating power adjustment quantities);

[0086] Dynamic correction: The actual values ​​of process parameters and quality indicators (gelatinization degree, hardness) are collected every 5 seconds and fed back to the BP neural network prediction model to update the prediction model parameters.

[0087] In S4, extrusion granulation uses a ring die extrusion granulator with a ring die diameter of 6-10mm.

[0088] S5. Online Quality Monitoring: Establish an online quality monitoring and feedback model. The online monitoring and feedback model collects particle spectral data through a near-infrared spectrometer and combines it with the PLSR model to detect the content of crude protein, crude fiber, and mycotoxins in real time.

[0089] The online monitoring and feedback model in S5 monitors key quality indicators of the finished feed product in real time, including crude protein. Coarse fiber Mycotoxins When the online monitoring and feedback model collects particle spectral data using a near-infrared spectrometer, a diffuse reflectance near-infrared spectrometer is installed at the outlet of the extrusion pellet mill to collect spectral data of the feed particles every 2 seconds. (dimension) (corresponding to absorbance at 1500 wavelength points), and then combined with the PLSR model to detect crude protein, crude fiber, and mycotoxin content in real time;

[0090] The PLSR model is used to establish a linear relationship between spectral data and quality indicators. The specific strategy is as follows:

[0091] C1. Sample set construction: Collect 1000 feed samples and measure their spectral data. Quality indicators compared with laboratory testing ;

[0092] C2. Data Preprocessing: [This section appears to be incomplete and requires further context.] Multivariate scattering correction, i.e., MSC and first derivative processing, is performed to eliminate interference caused by particle size and surface scattering.

[0093] C3.PLSR Modeling: Principal Component Extraction: Extraction via PLSR Each principal component, Dimensional reduction Establish a regression model : ,in The regression coefficient matrix , The residual matrix is ​​used for model validation, which involves cross-validation to obtain the coefficients of determination. Root mean square error This meets the monitoring accuracy requirements.

[0094] In S5, when online monitoring detects a quality indicator deviating from the target range, a feedback model is triggered. The specific strategy of the feedback model is as follows:

[0095] like Feedback to the multi-objective optimization model in S2 increases the proportion of protein raw materials such as soybean meal and fish meal, specifically by adjusting the amount. ,in This represents the actual crude protein value. This represents the target value for crude protein.

[0096] like The adaptive control model for processing parameters is fed back to S4 to improve the setpoint for tempering temperature. ;

[0097] like Immediately trigger an alarm, halt production, and investigate raw material contamination issues. This represents the actual mycotoxin value.

[0098] S6. Cooling and screening: The particles are cooled to room temperature by countercurrent cooling, and broken particles are removed by a grading sieve;

[0099] S7. Finished product packaging: Pack according to specifications, store in the warehouse, and the storage conditions are temperature ≤25℃ and relative humidity ≤60%.

[0100] Finally: The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A process for the production of high quality cattle and sheep feed, characterized in that, The specific steps are: S1. Raw material pretreatment: screening raw materials and removing impurities, drying at low temperature to 12-14% moisture, crushing to 1.0-1.5mm particle size, and the raw materials include but are not limited to corn, soybean meal, alfalfa, bran, premix; S2. Intelligent formula calculation: a multi-objective optimization model is established, the multi-objective optimization model includes objective functions with maximum nutrition satisfaction, minimum cost and maximum digestibility, the optimal raw material ratio is solved by improving the NSGA-II algorithm, and the constraint conditions of the multi-objective optimization model include raw material consumption, nutrition safety and physical property constraints; S3. Raw material mixing: according to the optimal ratio, put into the double-shaft paddle mixer, the mixing speed is 200r / min, the mixing time is 5min, and the uniformity coefficient is ≤7%; S4. Conditioning-extrusion granulation: an adaptive control model of processing parameters is established, and the adaptive control model of processing parameters adopts PID-neural network fusion control algorithm, and the conditioning temperature is 85-90℃, the extrusion speed is 280-320r / min, and the conditioning moisture is 14-16% according to the real-time moisture and particle size of the raw materials; S5. Online quality monitoring: an online quality monitoring and feedback model is established, and the online monitoring and feedback model collects particle spectrum data by near-infrared spectrometer, and combines PLSR model to detect crude protein, crude fiber and mycotoxin content in real time; S6. Cooling and screening: the particles are cooled to room temperature by countercurrent cooling, and the broken particles are removed by classification screen; S7. Finished product packaging: packaging according to specifications, storage, and storage conditions are temperature ≤25℃ and relative humidity ≤60%.

2. A process for the preparation of high quality cattle and sheep feed as claimed in claim 1, wherein: In S2, the objective function for maximizing nutritional satisfaction is used to measure the degree of matching between the key nutrients in the feed and the requirements of cattle and sheep. These key nutrients include, but are not limited to, crude protein (CP), crude fiber (CF), calcium (Ca), phosphorus (P), and vitamin A (AVA). Specifically, the objective function for maximizing nutritional satisfaction is as follows: , where m represents the key nutrient type; For the first The first of the raw materials The content of various nutrients; For the target cattle and sheep Requirements for various nutrients; For the first The quality percentage of each type of raw material; The closer the value is to 1, the more balanced the nutrition.

3. A process for the preparation of high quality cattle and sheep feed as claimed in claim 2, wherein: The objective function in S2 takes the raw material purchasing cost as the core, specifically: , is the unit cost of the first raw material. ​ 4. The process as claimed in claim 3, wherein the process is used for the production of high quality cattle and sheep feed. In S2, the objective function of maximizing nutrient digestibility is used to consider the synergistic effect of raw material combination on digestibility, and a digestibility synergy coefficient is introduced. Specifically: ,in For the first The first of the raw materials Basic digestibility of various nutrients; For the first , The combination of raw materials for the first The digestibility synergy coefficient of each nutrient, with a positive coefficient indicating synergistic enhancement and a negative coefficient indicating antagonism; For the first The weight of each nutrient.

5. A process for the preparation of high quality cattle and sheep feed as claimed in claim 4, wherein: The constraint conditions of the S2 multi-objective optimization model are specifically: Raw material usage constraints: wherein is the minimum addition amount of the th raw material; is the maximum addition amount of the th raw material; Nutritional safety constraints: ,in For the first Mycotoxin content in the raw materials For toxicity limits; Physical property constraints: ,in For the first The moisture content of the raw materials, This represents the upper limit of moisture content in the mixed raw materials.

6. A process for the preparation of high quality cattle and sheep feed as claimed in claim 1, wherein: In the improved NSGA-II algorithm in S2, an adaptive crossover and mutation operator is introduced, and the specific steps are: B1. Initialize population: generate individual raw material ratio schemes satisfying the constraint conditions; B2. Fast non-dominated sorting: calculate the non-dominated level and crowding degree of each individual, and select high-quality individuals; B3. Adaptive genetic operation: Crossover operator : where is the current iteration number, is the maximum iteration number, and a high crossover probability of 0.8 at the beginning of the iteration ensures diversity, and a low probability of 0.6 at the end promotes convergence. Variation operator : , high variation probability 0.2 to avoid local optimum after iteration B4. Elitist reserve strategy: Merge parents with offspring, select top individuals before breeding into next generation; B5. Termination condition: iteration to , output non-dominated level optimal solution set, according to user demand to select the final matching scheme.

7. A process for the preparation of high quality cattle and sheep feed as claimed in claim 6, wherein: The key parameters of the PID-neural network fusion algorithm control in the S4 include: conditioning temperature , extruder speed , conditioning moisture ; the process parameter control target is: starch gelatinization degree , particle hardness , and particle forming rate ; The structure of PID-neural network fusion algorithm is divided into prediction layer, i.e. BP neural network prediction model, and control layer, i.e. PID closed-loop control model, and the specific strategy is: The BP neural network prediction model is specifically: Input layer: 3 neurons, corresponding to raw material real-time moisture , raw material particle size , target gelatinization degree ; Hidden layer: 10 neurons, activation function is Sigmoid function; Output layer: 3 neurons corresponding to the optimal process parameter theoretical values , are the optimal conditioning temperature, the optimal extruder speed and the optimal conditioning moisture, respectively; Training procedure: Gradient descent method is used to minimize the prediction error , and wherein is the actual process parameter, is the predicted value, iterated to ; The PID closed-loop control model is specifically based on the output of the BP neural network prediction model. The PID setpoint is used to collect the actual values ​​of process parameters in real time. The control quantity is calculated using the positional PID formula, specifically: ,in ; For PID parameters; To control the quantity; Dynamic correction: collect process parameter actual value and quality index every 5s, feedback to BP neural network prediction model, and update prediction model parameters.

8. A process for the preparation of high quality cattle and sheep feed as claimed in claim 7, wherein: The line monitoring and feedback model in S5 monitors the key quality indexes of the feed product in real time, and the key quality indexes include crude protein , crude fiber , mycotoxin When the online monitoring and feedback model collects particle spectrum data by a near-infrared spectrometer, a diffuse reflection near-infrared spectrometer is arranged at the outlet of the extrusion granulator, and the spectrum data of the feed particles are collected every 2s Then, the content of crude protein, crude fiber and mycotoxin is detected in real time by combining a PLSR model. The PLSR model is used to establish the linear relationship between spectrum data and quality index, and the specific strategy is: C1. Sample set construction: 1000 feed samples were collected and spectral data were measured Quality indicators compared to laboratory testing ; C2. Data pre-processing: The data were subjected to multiple scattering correction (MSC) and first derivative treatment to eliminate the interference of particle size and surface scattering. C2. Data pre-processing: The data were subjected to multiple scattering correction (MSC) and first derivative treatment to eliminate the interference of particle size and surface scattering. C3. PLSR modeling: extract principal components: extract 3 principal components by PLSR reduce dimension to ; build regression model : , where is the regression coefficient matrix , is the residual matrix; model validation: cross-validation gives the determination coefficient , root mean square error , which meets the monitoring accuracy requirements.​ 9. A process for the preparation of high quality cattle and sheep feed as claimed in claim 8, wherein: When the quality index deviates from the target range in S5, the feedback model is triggered, and the specific strategy of the feedback model is: If : feedback to the multi-objective optimization model in S2, increase the proportion of protein raw materials such as soybean meal and fish meal, specifically: adjust the amount , wherein is the actual crude protein value; is the crude protein target value; If : Feedback to the process parameter adaptive control model in S4 to improve the set value of the conditioning temperature ; If : immediately trigger alarm, stop production, investigate raw material pollution problem, is the actual mycotoxin value.

10. The process as claimed in claim 8, wherein the process is used for the production of high quality cattle and sheep feed. In S4, the extrusion granulator adopts ring die extrusion granulator, and the ring die aperture is 6-10mm.

Images