Method for rapidly evaluating feed titer

Through modern chemical analysis and graph neural network-based model, the problem that the existing technology cannot evaluate feed titers in multiple aspects is solved, and a rapid and accurate feed titer assessment is achieved, which improves the growth and health level of animals.

CN118962011APending Publication Date: 2024-11-15JIANGSU INST OF POULTRY SCI +2
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411032545.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-30
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The prior art cannot quickly and accurately evaluate the titer of feed from multiple aspects, affecting the health and production performance of animals.

Method used

By quantitatively distributing feed samples, samples are collected, crushed, screened, flotation, decolorization, digestion and dissociation, and combined with modern chemical analysis methods, the nutrient components and moisture in the feed are measured, and the conversion and digestibility of the feed is calculated. The state transfer model and reinforcement learning strategy optimization of graph neural networks evaluate the titer of feed.

Benefits of technology

It improves the accuracy and speed of feed evaluation, can provide animals with high-quality feed choices, reduce feeding costs, and promote animal growth and health.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118962011A_ABST
    Figure CN118962011A_ABST
Patent Text Reader

Abstract

The invention discloses a method for rapidly evaluating feed titer, relates to the technical field of feed titer evaluation, and solves the problems that the feed titer cannot be evaluated from different aspects, the accuracy of feed titer evaluation is influenced, errors are easy to occur when proper feed is selected by animals, and the feed titer evaluation accuracy is influenced. The health level and the production performance of animals are influenced. According to the method, the nutrient substance content, the absolute titer, the relative titer, the digestibility determination and the feed conversion rate of the feed are comprehensively considered, so that the feed titer can be evaluated in multiple aspects, the accuracy of evaluating the feed titer can be improved, the feed titer can be quickly and accurately evaluated, high-quality feed selection can be provided for the breeding industry, and the economic benefit is increased. The feed additive has important significance for improving the animal production efficiency, promoting the growth and health of animals and reducing the feeding cost.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of evaluating feed potency, in particular to a method for rapidly evaluating feed potency. Background Art

[0002] Feed potency refers to the amount of effective ingredients contained in feed, which has an important impact on the growth and health of animals. The higher the feed potency, the more it can promote the growth of animals and reduce feed costs. On the contrary, low feed potency may lead to slower growth of animals and a corresponding increase in feeding costs. Feed is the material basis for animal growth and production, and its potency directly affects the health level and production performance of animals. Therefore, quickly and accurately evaluating feed potency helps provide animals with suitable feed.

[0003] The existing method for evaluating feed potency is single and cannot evaluate feed potency from different aspects, which affects the accuracy of feed potency evaluation and easily leads to errors when animals choose suitable feed, affecting the health level and production performance of animals. Therefore, a method for quickly evaluating feed potency is proposed. Summary of the invention

[0004] The object of the present invention is to provide a method for rapidly evaluating feed potency to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for rapidly evaluating feed potency comprises the following steps:

[0007] The feed to be measured is placed on a weighing machine, and based on the weighing result, the weight of the feed to be measured is determined, and the operation is repeated to quantitatively distribute the feed sample into multiple portions;

[0008] The feed to be tested is sampled, crushed, sieved, floated, decolorized, digested and dissociated, and mounted on a slide, and then observed under a stereo microscope to determine the nutritional components of the feed to be tested;

[0009] Based on the determined nutritional components of the feed to be tested, combined with the measured weight of the feed, the nutritional components in the feed are quantitatively analyzed, and the nutritional components include crude protein CP, ether extract EE, crude fiber CF, ash Ash, water DM, nitrogen-free extract NFE and trace elements, and the trace elements include calcium, phosphorus, salt, and vitamins;

[0010] According to the results of feed nutrient analysis, the content and proportion of various nutrients in the feed are evaluated.

[0011] Moisture DM content: Put the quantitative feed sample into the moisture detector to measure the moisture and volatile matter to obtain the moisture and volatile matter content;

[0012] Ash content: Put the quantitative feed sample into the burner and burn it at 500-600℃ for 2h to get the mineral content; Mineral energy (%) = (mineral content / daily recommended intake) × 100%;

[0013] Crude protein CP content: Put the feed sample into a container, digest the feed sample with concentrated sulfuric acid to convert all organic nitrogen into inorganic ammonium salts, then convert the ammonium salts into ammonia under alkaline conditions, distill it out with water vapor and absorb it with excess boric acid solution, then titrate it with standard hydrochloric acid to calculate the nitrogen content in the sample. The nitrogen content in protein usually accounts for 16% of its total mass. Therefore, by measuring the nitrogen content in the substance, the total protein content in the substance can be estimated, that is: protein content = nitrogen content / 16%; protein energy (%) = (protein content × 4 kcal / g) / total energy × 100%;

[0014] Ether extract EE content: Mix the feed sample with ether and stir to dissolve the fat in the ether, then separate the solution from the residue by filtering or centrifuging, and finally evaporate the ether by heating. The remaining residue is the fat content of the feed sample; Fat energy (%) = (fat content × 9 kcal / g) / total energy × 100%;

[0015] Crude fiber CF content: Put the feed sample into weak acid and weak alkali solution and boil for 30 minutes. Use weak acid solution and weak alkali solution to destroy or dissolve some non-cellulose substances that are closely bound to cellulose. The boiling process helps to accelerate the chemical reaction and make the treatment more thorough. Then the treated solution is separated into solid and liquid to obtain the crude fiber residue after acid and alkali treatment, that is, the crude fiber content. The calculation formula is: crude fiber content (%) = (burning residue mass / feed sample mass) × 100%; crude fiber CF energy (%) = (crude fiber CF content / daily recommended intake) × 100%;

[0016] NFE content of nitrogen-free extract: NFE% = 100% - (DM% + CP% + EE% + CF% + Ash%);

[0017] Carbohydrate energy (%) = (carbohydrate content × 4 kcal / g) / total energy × 100%

[0018] Calcium content: destroy the organic matter in the feed sample, convert calcium into water-soluble ions, quantitatively precipitate with ammonium oxalate, and indirectly determine the calcium content using the potassium permanganate method;

[0019] Phosphorus content: First, the organic matter in the feed sample is destroyed to release phosphorus, and then it is treated with ammonium vanadium molybdate in an acidic solution to produce yellow (NH4+)3PONH4+VO3·16MoO3. The colorimetric determination is performed at a wavelength of 420nm to measure the total phosphorus content, which includes phytic phosphorus that is difficult for animals to absorb.

[0020] Salt content: Under acidic conditions, add excess silver nitrate solution to make the chloride in the sample form silver chloride precipitation. After removing the precipitation, use ammonium thiocyanate to back-titrate the excess silver nitrate. Calculate the chloride content based on the amount of ammonium thiocyanate consumed.

[0021] Vitamin content: Heat and saponify the feed sample with potassium hydroxide-ethanol solution to release the vitamins, then extract with organic solvents and analyze and determine with high performance liquid chromatography to achieve qualitative and quantitative analysis of multiple vitamins in the feed. During the sample processing, the antioxidant 2,6-di-tert-butyltoluene (BHT) needs to be added for protection;

[0022] According to the results of feed component analysis, the content and proportion of various nutrients in the feed are evaluated. According to the nutritional requirements of different animals, it is analyzed whether the feed meets the needs of animal growth and production, so as to evaluate the potency of the feed;

[0023] The absolute potency and relative potency of feed are also important indicators for evaluating feed potency. The absolute potency reflects the content of active ingredients in the feed, while the relative potency compares the difference between the feed sample and the standard feed sample.

[0024] Based on the content and energy of various nutrients in the feed, calculate the ratio of animal weight gain to feed consumption and evaluate the feed conversion rate;

[0025] Based on the content and energy of various nutrients in the feed, the digestibility of each nutrient in the feed is determined by collecting and analyzing animal feces.

[0026] Preferably, the method for calculating the ratio of animal weight gain to feed consumption and evaluating feed conversion rate is:

[0027] Record the initial body weight W1 of the animal and the final body weight W2 after feeding for a period of time, and calculate the weight gain G of the animal: G = W1-W2;

[0028] Record the total amount of feed consumed by the animal during the specified time period F, in kilograms;

[0029] Calculate the feed conversion ratio (CR): CR = G / F;

[0030] Among them, CR is the feed conversion rate, which indicates the animal weight gain (kg) that can be achieved per kilogram of feed consumption.

[0031] Preferably, the method for collecting and analyzing animal feces is:

[0032] The pre-feeding period for metabolic testing of poultry should be no less than 4 days, and feces should be collected continuously for 3 days afterwards;

[0033] Cages should be used for raising poultry, and clean, easy-to-collect feces collection devices, such as feces collection trays, plastic films or kraft paper, should be placed under the cages to accurately and completely collect poultry excrement. The collected feces samples need to be properly processed, and impurities such as feed, feathers and dandruff that contaminate the samples should be picked out. The collected feces samples should be dried, ground, sieved and stored;

[0034] Use chemical analysis methods, such as spectral analysis, chromatography, etc., to determine the content of various nutrients in fecal samples. These nutrients may include protein, fat, cellulose, minerals and vitamins, etc.; by comparing the content of nutrients in feed and feces, the digestibility of feed can be calculated.

[0035] Preferably, based on the determination of the content of various nutrients in the fecal sample, the method for evaluating the real available energy of the feed in the animal body is:

[0036] Digestible energy DE can be divided into apparent digestible energy ADE and true digestible energy TDE;

[0037] ADE refers to the total energy of feed minus the energy in feces, that is, the undigested energy in feed, including the excretions and secretions from the animal's digestive tract, and the endogenous energy produced by intestinal microorganisms;

[0038] TDE is the apparent metabolizable energy plus the endogenous energy value normally produced in the animal body;

[0039] Metabolizable energy ME refers to the energy of feed nutrients that participate in the transformation of animal bodies, that is, the remaining energy after deducting fecal energy, urine energy and methane energy from the total energy value of feed. ME can be divided into apparent metabolizable energy AME and true metabolizable energy TME. TME is AME plus endogenous energy in urine and feces.

[0040] In order to compare the metabolizable energy values ​​of different feeds, the influence of nitrogen deposition on the ME value should be eliminated, that is, the metabolizable energy should be corrected according to the nitrogen deposition to make it the ME when the nitrogen deposition is zero. The correction formula is:

[0041] AMEn=AME-RN×34.39

[0042] TMEn=TME-RN×34.39

[0043] Where: RN (Total Nitrogen Retained) is the amount of nitrogen deposited by poultry per day (g), which can be positive, negative or zero. The sign is substituted in the calculation. 34.39 is the energy corresponding to each gram of urinary nitrogen;

[0044] TME stands for True Metabolizable Energy, which is the energy actually absorbed and utilized by the animal body, that is, the energy after deducting fecal energy, urine energy and endogenous nitrogen loss in the digestive tract. The calculation formula is generally: TME = TDE-[(UE-UeE)+Eg]

[0045] Among them, TDE is total digestible energy, UE is urine energy, UeE is endogenous urine energy, and Eg is energy loss of endogenous nitrogen in the digestive tract;

[0046] The relationship between metabolizable energy, digestible energy and crude protein is:

[0047] ME=DE×[(96-0.202×CP) / 100]

[0048] That is, for every 1 percentage point increase in crude protein, the utilization rate of digestible energy converted into metabolic energy decreases by 0.202 percentage points.

[0049] Compared with the prior art, the invention has the following beneficial effects: the method for rapidly evaluating feed potency quantitatively analyzes the nutrients in the feed by means of modern chemical analysis, including proteins, fats, fibers, minerals, vitamins and other components which are necessary for the growth of livestock and poultry, the content and proportion of the nutrients will directly affect the nutritional value of the feed, and the feed potency is evaluated by measuring the nutrients in the feed; the digestibility of each nutrient in the feed is measured by collecting and analyzing animal feces, the digestibility directly reflects the utilization degree of the nutrients in the feed, and the feed with high digestibility means that the nutrients therein are more easily absorbed and utilized by livestock and poultry, thereby improving the potency of the feed; the conversion rate of the feed is evaluated by calculating the ratio of the weight gain of the animal to the feed consumption, and the higher the conversion rate, the higher the feed potency; by comprehensively considering the nutrient content, absolute potency, relative potency, digestibility determination and feed conversion rate of the feed, the feed potency is evaluated in multiple aspects, the accuracy of evaluating the feed potency can be improved, the feed potency can be evaluated quickly and accurately, and the high-quality feed selection can be provided for the breeding industry, which is of great significance for improving animal production efficiency, promoting animal growth and health, and reducing breeding costs.

[0050] Comprehensiveness of data collection: The present invention systematically collects multiple nutritional data and animal growth data of feed samples, including crude protein (CP), ether extract (EE), crude fiber (CF), ash (Ash), moisture (DM), nitrogen-free extract (NFE), and trace elements such as calcium, phosphorus, salt, vitamins, etc. These data provide a comprehensive and accurate basis for subsequent analysis and modeling, ensuring that the model can accurately reflect the impact of feed ingredients on animal growth.

[0051] Scientificity of data preprocessing: Through data cleaning, missing value processing, outlier detection and processing, and standardization processing, the present invention ensures the integrity, accuracy, and consistency of the data. Specific methods include using the mean to fill missing values, z-score outlier detection, and median replacement, which effectively reduce the interference of noise data on the model and improve data quality.

[0052] Innovation of feature engineering: The present invention enhances the expressiveness of data through feature engineering, including generating polynomial features (such as the product of CP and EE), interactive features (such as the ratio of CP / EE), and time series features (considering the relationship between the components of the previous feedings and the current components). These feature engineering methods can capture the complex relationships and nonlinear effects between feed components, and improve the predictive performance and stability of the model.

[0053] State transition model based on graph neural network: The present invention adopts a state transition model based on graph neural network (GNN), which takes into account the relationship and dependency between components. GNN captures the complex relationship in the data through nodes (representing feed components) and edges (representing the relationship between components), and recursively propagates node information to adjacent nodes to achieve information integration and feature extraction. This method can effectively capture complex relationships and nonlinear effects, providing more accurate predictions.

[0054] Reasonable design of reward function: The reward function designed by the present invention reflects the animal weight gain and feed utilization efficiency, which is defined as the ratio of animal weight gain to feed consumption. By measuring animal weight gain and feed consumption, the reward function can accurately evaluate the effect of each formula. The efficient reward function design guides the model to make optimal decisions, improving the potency of feed and the health of animals.

[0055] Efficiency of reinforcement learning strategy optimization: The present invention uses the PPO algorithm for strategy optimization, and ensures the stability and efficiency of the training process by updating the policy network and the value network. The specific steps include initializing the policy network and the value network, selecting actions from the policy network, storing experience data, and updating the network. The detailed calculation and optimization process ensures the performance and stability of the model.

[0056] Significance of overall optimization effect: Through comprehensive data collection and preprocessing, innovative feature engineering, advanced state transition model and reinforcement learning strategy optimization, the present invention can significantly improve the optimization effect of feed formula. Ultimately, it improves the utilization efficiency of feed and the growth and health level of animals, which has obvious advantages and improvements compared with the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A schematic diagram of a method for rapidly evaluating feed potency according to the present invention;

[0058] Figure 2 A schematic diagram of a feed conversion rate process of a method for rapidly evaluating feed potency of the present invention;

[0059] Figure 3 A schematic diagram of the animal feces collection and analysis process of a method for rapidly evaluating feed potency of the present invention;

[0060] Figure 4 The present invention is a flow chart of an intelligent algorithm introduced into a method for rapidly evaluating feed potency. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] Example

[0063] See also Figure 1-3 , the present invention provides an embodiment: a method for quickly evaluating feed potency, comprising the following steps:

[0064] The feed to be measured is placed on a weighing machine, and based on the weighing result, the weight of the feed to be measured is determined, and the operation is repeated to quantitatively distribute the feed sample into multiple portions;

[0065] The feed to be tested is sampled, crushed, sieved, floated, decolorized, digested and dissociated, and mounted on a slide, and then observed under a stereo microscope. The morphological characteristics, physicochemical characteristics, and physical form of the feed under the microscope are compared and analyzed with the characteristics of the actual feed raw materials to determine the nutritional components of the feed to be tested;

[0066] Based on the determined nutritional components of the feed to be tested, combined with the measured weight of the feed, the nutritional components in the feed are quantitatively analyzed, the nutritional components include crude protein CP, ether extract EE, crude fiber CF, ash Ash, water DM, nitrogen-free extract NFE and trace elements, the trace elements include calcium, phosphorus, salt, vitamins, crude protein CP includes: protein, amino acids, non-protein nitrogen, ether extract EE includes: fat, oil, wax, pigment, resin, crude fiber CF includes: cellulose, hemicellulose, lignin, ash Ash includes minerals, water DM includes water and volatile substances, nitrogen-free extract NFE includes: starch, sugar, part of cellulose, hemicellulose, lignin;

[0067] According to the results of feed nutrient analysis, the content and proportion of various nutrients in the feed are evaluated.

[0068] Moisture DM content: Put the quantitative feed sample into the moisture detector to measure the moisture and volatile matter to obtain the moisture and volatile matter content;

[0069] Ash content: Put a quantitative feed sample into a burner and burn it at 500-600℃ for 2h to obtain the mineral content; it is obtained by comparing its content in food or feed with the recommended intake or reference intake, and the formula is: Mineral energy (%) = (mineral content / daily recommended intake) × 100%;

[0070] Crude protein CP content: Put the feed sample into a container, digest the feed sample with concentrated sulfuric acid to convert all organic nitrogen into inorganic ammonium salts, then convert the ammonium salts into ammonia under alkaline conditions, distill it out with water vapor and absorb it with excess boric acid solution, then titrate it with standard hydrochloric acid to calculate the amount of nitrogen in the sample. The nitrogen content in protein usually accounts for 16% of its total mass. Therefore, by measuring the nitrogen content in the substance, the total protein content in the substance can be estimated (assuming that all the nitrogen in the measured substance comes from protein), that is: protein content = nitrogen content / 16%; protein energy (%) = (protein content×4 kcal / g) / total energy×100%

[0071] Ether extract EE content: Mix the feed sample with ether to dissolve the fat in the ether, then separate the solution from the residue by filtering or centrifuging, and finally evaporate the ether by heating. The remaining residue is the fat content of the feed sample; Fat energy (%) = (fat content × 9 kcal / g) / total energy × 100%

[0072] Crude fiber CF content: Put the feed sample into weak acid and weak alkali solution and boil for 30 minutes. Use weak acid solution and weak alkali solution to destroy or dissolve some non-cellulose substances that are closely bound to cellulose. The boiling process helps to accelerate the chemical reaction and make the treatment more thorough. Then the treated solution is separated into solid and liquid to obtain the crude fiber residue after acid and alkali treatment, that is, the crude fiber content. The calculation formula is: crude fiber content (%) = (burning residue mass / feed sample mass) × 100%; crude fiber CF energy (%) = (crude fiber CF content / daily recommended intake) × 100%;

[0073] NFE content of nitrogen-free extract: NFE% = 100% - (DM% + CP% + EE% + CF% + Ash%);

[0074] Carbohydrate energy (%) = (carbohydrate content × 4 kcal / g) / total energy × 100%

[0075] Calcium content: destroy the organic matter in the feed sample, convert calcium into water-soluble ions, quantitatively precipitate with ammonium oxalate, and indirectly determine the calcium content using the potassium permanganate method;

[0076] Phosphorus content: First, the organic matter in the feed sample is destroyed to release phosphorus, and then it is treated with ammonium vanadium molybdate in an acidic solution to produce yellow (NH4+)3PONH4+VO3·16MoO3. The colorimetric determination is performed at a wavelength of 420nm to measure the total phosphorus content, which includes phytic phosphorus that is difficult for animals to absorb.

[0077] Salt content: Under acidic conditions, add excess silver nitrate solution to make the chloride in the sample form silver chloride precipitation. After removing the precipitation, use ammonium thiocyanate to back-titrate the excess silver nitrate. Calculate the chloride content based on the amount of ammonium thiocyanate consumed.

[0078] Vitamin content: Heat and saponify the feed sample with potassium hydroxide-ethanol solution to release the vitamins, then extract with organic solvents and analyze and determine with high performance liquid chromatography to achieve qualitative and quantitative analysis of multiple vitamins in the feed. During the sample processing, the antioxidant 2,6-di-tert-butyltoluene (BHT) needs to be added for protection;

[0079] According to the results of feed component analysis, the content and proportion of various nutrients in the feed are evaluated. According to the nutritional requirements of different animals, it is analyzed whether the feed meets the needs of animal growth and production, so as to evaluate the potency of the feed;

[0080] The absolute potency and relative potency of feed are also important indicators for evaluating feed potency. Relative potency is the ratio of the required content of a known feed sample to that of a standard feed sample. Its unit is usually a relative unit or percentage. For example, a relative potency of a feed sample of 90% means that the active ingredient contained in the sample is 10% lower than that of the standard feed. Absolute potency is the ratio of the content unit of active ingredients such as vitamins and minerals in the feed to the content unit required by the feed. These indicators can quantitatively describe the content and proportion of active ingredients in the feed. Absolute potency reflects the content of active ingredients in the feed, while relative potency compares the difference between the feed sample and the standard feed sample.

[0081] Based on the content and energy of various nutrients in the feed, the ratio of animal weight gain to feed consumption is calculated to evaluate the feed conversion rate. The method for calculating the ratio of animal weight gain to feed consumption and evaluating the feed conversion rate is as follows:

[0082] Record the initial body weight W1 of the animal, in kilograms, and the final body weight W2 after a period of feeding, in kilograms, and calculate the weight gain G of the animal: G = W1-W2;

[0083] Record the total amount of feed consumed by the animal during the specified time period F, in kilograms;

[0084] Calculate the feed conversion ratio (CR): CR = G / F;

[0085] Among them, CR is the feed conversion rate, which indicates the animal weight gain (kg) that can be achieved per kilogram of feed consumption;

[0086] Assuming that the animal's weight gain and feed consumption are known, calculate its feed conversion rate. The result is: conversion rate = 1.25. Therefore, the animal's feed conversion rate is: 1.25, which means that for every 1 kg of feed consumed by the animal, it can gain 1.25 kg of body weight. The higher the animal's weight gain, the higher the conversion rate. The higher the conversion rate, the higher the feed efficacy, which is more conducive to reducing feeding costs.

[0087] By calculating the feed conversion rate, we can evaluate the effects of different feeds on animal growth, so as to select the most suitable feed for animal growth and improve feeding efficiency;

[0088] Based on the content and energy of various nutrients in feed, the digestibility of each nutrient in feed is determined by collecting and analyzing animal feces. The digestibility directly reflects the utilization degree of nutrients in feed, and the utilization degree of nutrients in feed reflects the feed potency. The method of collecting and analyzing animal feces is:

[0089] The pre-feeding period for metabolic testing of poultry should be no less than 4 days, and feces should be collected continuously for 3 days afterwards;

[0090] Cages should be used for raising poultry, and clean, easy-to-collect feces collection devices, such as feces collection trays, plastic films or kraft paper, should be placed under the cages to accurately and completely collect poultry excrement. The collected feces samples need to be properly processed, and impurities such as feed, feathers and dandruff that contaminate the samples should be picked out. The collected feces samples should be dried, ground, sieved and stored;

[0091] Use chemical analysis methods, such as spectral analysis, chromatography, etc., to determine the content of various nutrients in fecal samples. These nutrients may include protein, fat, cellulose, minerals and vitamins. By comparing the content of nutrients in feed and feces, the digestibility of feed can be calculated.

[0092] The method for evaluating the true available energy of feed in the animal is based on the determination of the content of various nutrients in fecal samples:

[0093] Digestible energy DE can be divided into apparent digestible energy ADE and true digestible energy TDE;

[0094] ADE refers to the total energy of feed minus the energy in feces, that is, the undigested energy in feed, including the excretions and secretions from the animal's digestive tract, and the endogenous energy produced by intestinal microorganisms;

[0095] TDE is the apparent metabolizable energy plus the endogenous energy value normally produced in the animal body;

[0096] Metabolizable energy ME refers to the energy of feed nutrients that participate in the transformation of animal bodies, that is, the remaining energy after deducting fecal energy, urine energy and methane energy from the total energy value of feed. ME can be divided into apparent metabolizable energy AME and true metabolizable energy TME. TME is AME plus endogenous energy in urine and feces.

[0097] In order to compare the metabolizable energy values ​​of different feeds, the influence of nitrogen deposition on the ME value should be eliminated, that is, the metabolizable energy should be corrected according to the nitrogen deposition to make it the ME when the nitrogen deposition is zero. The correction formula is:

[0098] AMEn=AME-RN×34.39

[0099] TMEn=TME-RN×34.39

[0100] Where: RN (Total Nitrogen Retained) is the amount of nitrogen deposited by poultry per day (g), which can be positive, negative or zero. The sign is substituted in the calculation. 34.39 is the energy corresponding to each gram of urinary nitrogen;

[0101] TME stands for True Metabolizable Energy, which is the energy actually absorbed and utilized by the animal body, that is, the energy after deducting fecal energy, urine energy and endogenous nitrogen loss in the digestive tract. The calculation formula is generally: TME = TDE-[(UE-UeE)+Eg]

[0102] Among them, TDE is total digestible energy, UE is urine energy, UeE is endogenous urine energy, and Eg is energy loss of endogenous nitrogen in the digestive tract;

[0103] The relationship between metabolizable energy, digestible energy and crude protein is:

[0104] ME=DE×[(96-0.202×CP) / 100]

[0105] That is, for every 1 percentage point increase in crude protein, the utilization rate of digestible energy converted into metabolic energy decreases by 0.202 percentage points.

[0106] like Figure 4 Based on the determined nutritional components of the feed to be measured, combined with the measured weight of the measured feed, the nutritional components in the feed are quantitatively analyzed, and a model-based reinforcement learning intelligent algorithm is introduced to optimize the process of quantitative analysis of the nutritional components in the feed. The specific process is:

[0107] Step 1: Data collection and preprocessing

[0108] Data collection: Collect nutritional data of a large number of feed samples and animal growth data for subsequent analysis and modeling; specifically includes the following components:

[0109] Crude protein (CP): Protein is the main nutrient required for animal growth and repair of body tissues; crude protein is the sum of all protein components in feed, including protein from animal and plant sources; protein deficiency can lead to slow growth and decreased immunity in animals, so it is an important indicator for evaluating feed quality;

[0110] Ether extract (EE): refers to the total amount of fats and oils in feed extracted by ether; these ingredients provide animals with high energy and help absorb fat-soluble vitamins, vitamins A, D, E and K; the appropriateness of fat content directly affects the energy intake and balance of animals;

[0111] Crude fiber (CF): refers to the acid- and alkali-insoluble cellulose part of feed, mainly including cellulose, hemicellulose and lignin; crude fiber is essential to the digestive health of animals, can promote intestinal peristalsis, prevent constipation, and help regulate the function of the digestive system;

[0112] Ash: refers to the total amount of minerals in feed, including calcium, phosphorus, and magnesium. Minerals are very important for the development of bones and teeth and metabolic functions of animals. Calcium and phosphorus are the main components of bone development, while magnesium is involved in many metabolic processes.

[0113] Moisture (DM): refers to the moisture content in feed; moisture content affects the storage and use of feed. Too much moisture causes the feed to mold, while too little moisture affects the palatability and digestibility of the feed.

[0114] Nitrogen-free extract (NFE): refers to carbohydrates, starches and sugars in feed that do not contain nitrogen; these ingredients provide energy for animals and are an important source of energy for animal activity and growth;

[0115] Trace elements: including calcium, phosphorus, salt, vitamins, these elements in small amounts are essential for the growth, development and health of animals; although trace elements and vitamins are required in small quantities, their deficiency can cause serious health problems;

[0116] Data preprocessing:

[0117] Data cleaning:

[0118] Check for missing values: For each sample, check whether there are missing values; missing values ​​are incomplete data due to measurement errors or recording errors; identify these missing values ​​to ensure the completeness and accuracy of the data;

[0119] Handling missing values: Missing values ​​can be filled with the mean to maintain the overall trend and distribution of the data;

[0120] Mean filling formula:

[0121]

[0122] Among them, n new : The number of new data; x i is a non-missing value, and n is the number of non-missing values; deleting samples with missing values ​​can avoid the impact of incomplete data on the analysis results, but it will reduce the amount of data;

[0123] Check outliers: Use the z-score method to identify values ​​far from the mean. Outliers are extreme values ​​caused by input errors or equipment failures. Values ​​exceeding 3 standard deviations are usually considered outliers.

[0124] The z-score formula is:

[0125]

[0126] Among them, z is the recognition criterion, x is the sample value, μ is the mean, and σ is the standard deviation;

[0127] Handling outliers: Use the median replacement method to handle outliers. Median replacement can reduce the impact of outliers on data distribution, thereby improving the reliability of analysis results. The median replacement method effectively reduces the impact of outliers on data because the median is less affected by extreme values.

[0128] Median replacement formula:

[0129] x new =median(x 1 ,x 2 ,...,x n )

[0130] x new : New data value, that is, the data value after replacing abnormal values ​​or missing values;

[0131] (x 1 ,x 2 ,...,x n ): data values ​​in the original data set, which do not contain outliers or missing values; that is, only normal, reasonable data points are used to calculate the median;

[0132] median(x 1 ,x 2 ,...,x n ): The median of the original data set; the median is the value in the middle after all the data are arranged in order of size; for an odd number of data points, the median is the middle number; for an even number of data points, the median is the average of the two middle numbers;

[0133] Standardization:

[0134] Each nutrient component data is standardized so that its mean is 0 and its standard deviation is 1. The purpose of standardization is to convert data of different dimensions to the same scale so that each component contributes equally to the model. This helps speed up the model training process and improves the performance of the model. After standardization, the distribution of each component data is more consistent, which is conducive to the rapid convergence and stability of the algorithm.

[0135]

[0137] Among them, x` is the numerical value corresponding to each data value after standardization, x is the sample value, μ is the mean, and σ is the standard deviation;

[0138] Feature Engineering:

[0139] Introducing feature engineering to enhance the expressiveness of data by combining, selecting or generating new features; feature engineering can mine more useful information and improve the predictive performance of the model;

[0140] Polynomial features: Generate polynomial features between the components; Generate the product feature of CP and EE to represent the interaction between these two components; By generating polynomial features, the nonlinear relationship between feed components can be captured;

[0141] Formula: NewFeature=q 1 ×q 2

[0142] NewFeature: New feature

[0143] q i : An original characteristic, any two of crude protein CP, ether extract EE, crude fiber CF, ash Ash, moisture DM, nitrogen-free extract NFE, and trace elements;

[0144] Interaction features: Combine the interaction features between components; generate CP / EE ratio features to represent the relative relationship between the two components; reveal the relative importance and mutual influence of feed components;

[0145] Time series characteristics: Consider the characteristics of time series data, the relationship between the ingredients of the previous feedings and the current ingredients; this can help the model capture the cumulative impact of feed ingredients on animal growth; reflect the changing trend of feed ingredients over time and the long-term impact on animal growth;

[0146] formula:

[0147] NewFeature t =f(NewFeature t-1 ,NewFeature t-2 ,...,NewFeature t-n )

[0148] Among them, NewFeature t : The new feature at time t; this is the feature value calculated by function f based on the feature values ​​of the previous n time steps;

[0149] f: Function f represents the mathematical relationship or algorithm for converting the eigenvalues ​​of the previous n time steps into the new eigenvalues ​​of the current time step. This function can be in the form of weighted average, cumulative sum, or difference.

[0150] NewFeature t-1 : The eigenvalue at time t-1; represents the eigenvalue at one time step before the current time step;

[0151] Step 2: Environmental model construction

[0152] State representation: define state st The nutritional components of the current feed formula and the growth status of the animal; the state vector contains the specific values ​​of different components and the weight information of the animal;

[0153] Example of a state vector:

[0154] s t =[CP,EE,CF,Ash,DM,NFE,Ca,P,Salt,Vitamin,Weight]

[0155] Specific explanation: Ca: calcium, an important mineral element; P: phosphorus, an important mineral element; Salt: salt, an important component for maintaining electrolyte balance; Vitamin: vitamin, a micronutrient that supports growth and immune function; Weight: animal weight, indicating the current weight of the animal;

[0156] Action representation: define action a t To make decisions on adjusting feed formulation, it means increasing or decreasing the proportion of a certain ingredient;

[0157] Example of action vector:

[0158] a t =[ΔCP,ΔEE,ΔCF,ΔAsh,ΔDM,ΔNFE,ΔCa,ΔP,ΔSalt,ΔVitamin]

[0159] Specific explanation: ΔCP: adjustment of crude protein; ΔEE: adjustment of ether extract; ΔCF: adjustment of crude fiber; ΔAsh: adjustment of ash; ΔDM: adjustment of water; ΔNFE: adjustment of nitrogen-free extract; ΔCa: adjustment of calcium; ΔP: adjustment of phosphorus; ΔSalt: adjustment of salt; ΔVitamin: adjustment of vitamins;

[0160] State transfer model: It uses a graph neural network-based approach to consider the relationships and dependencies between components. The graph neural network model can effectively capture complex relationships and nonlinear effects, providing more accurate predictions.

[0161] State transition formula

[0162]

[0163] Specific explanation:

[0164] s t+1 : represents the new state at time t+1, that is, when executing action a t Afterwards, changes in feed composition and animal weight;

[0165] The input of the graph neural network model is the current state s tand action a t , the output is the new state s t+1 ;

[0166] GNN; graph neural network function;

[0167] Input layer: current state s t and action a t As input; each input node represents a component of the state vector;

[0168] Graph construction: Build a graph structure containing all components and animal weights, and the relationships between components are connected by edges;

[0169] Feature extraction: The features of each node (component) are extracted through a multi-layer graph neural network model layer; the graph neural network model layer aggregates the information of each node with the information of its adjacent nodes through a neighborhood aggregation operation to form a new feature representation;

[0170] Output layer: output new state s t+1 , i.e., the changes in each component and the animal's weight after the action was performed;

[0171] Mathematical expression of graph neural network model:

[0172] Each layer of the graph neural network model updates its features using the following formula:

[0173]

[0174] Where: v: node; u: neighbor node; The initial features represent the features of node v in the kth layer; represents the characteristics of the k-th layer neighbor node u; N(v) represents the set of neighbor nodes of node v; W (k) The weight matrix of the kth layer; linearly transform the information of neighbor nodes and then aggregate them; this matrix is ​​a parameter that needs to be learned through training; b (k) The bias vector of the kth layer; the bias term is introduced to improve the expressiveness of the model; this vector is also a parameter that needs to be learned through training; σ is a nonlinear activation function; the introduction of nonlinearity enables the model to capture complex relationships; common activation functions include ReLU, Sigmoid, and Tanh;

[0175] Finally, after passing through several layers of graph neural network models, a new feature representation of each node is obtained, that is, the new state s t+1 ;

[0176] Step 3: Reward Function Design

[0177] Reward function r tIt is a key component used in reinforcement learning to evaluate the effect of an action under a specific state. In the problem of optimizing animal feed formula, the design of the reward function should reflect the weight gain effect of the animal and the efficiency of feed utilization. An effective reward function can guide the model to make optimal decisions, improve the potency of the feed and the health of the animal.

[0178] The reward function can be defined as the ratio of animal weight gain to feed consumption

[0179]

[0180] Specific explanation

[0181] r t : The reward value at time t reflects the animal weight gain and feed utilization efficiency; the higher the reward value, the higher the feed utilization efficiency and the better the animal weight gain effect; a high reward value can guide the model to make optimal decisions and improve the potency of feed and the health of animals;

[0182] ΔWeight t : Animal weight gain at time t, indicating the change in animal weight; weight gain is an important indicator for evaluating the effect of feed formula;

[0183] FeedConsumption t : Feed consumption at time t, indicating the feed intake of the animal in a specific time period;

[0184] Step 4: Reinforcement Learning Strategy Optimization

[0185] Strategy representation: Define a strategy π(a t |s t ) is in state s t Next select action a t The probability of the policy network is used to approximate this policy, and the input is the current state s t , output different actions a t The probability distribution of

[0186] This means that in a given state, the policy network will calculate the probability of selecting each possible action, thereby guiding the decision-making process;

[0187] Algorithm selection: Use the model-based reinforcement learning algorithm PPO; PPO is an advanced reinforcement learning algorithm that maintains the stability and efficiency of the training process by limiting the amplitude of strategy updates and avoiding instability caused by drastic changes in strategies;

[0188] initialization

[0189] Policy network π(s t |θ π): used to output the probability distribution of each action in each state; here θ π Represents the parameters of the policy network; these parameters are optimized through training so that the policy network can correctly estimate the probability of the best action;

[0190] Value network V(s t |θ V ): used to estimate the value function of each state, that is, in state s t The expected return under V Represents the parameters of the value network; the value network helps the policy network evaluate the potential value of each state;

[0191] Experience replay buffer R: used to store the experience data of each step, including state, action, reward and next state; these experience data will be used to update network parameters in subsequent training;

[0192] Action Selection

[0193] At each time step, an action a is selected from the policy network t =π(s t |θ π ); the specific process is as follows:

[0194] Enter the current state s t To the policy network, output the probability distribution of each action;

[0195] Sampling is performed based on the output probability distribution, and action a is selected t ; This means that if the probability of an action is high, the probability of it being selected will also be high;

[0196] Storage experience

[0197] The experience (s t ,a t ,r t ,s t+1 ) is stored in the experience replay buffer R; these experience data are used for subsequent network updates;

[0198] Update Network

[0199] Randomly sample mini-batches (s) from the experience replay buffer t ,a t ,r t ,s t+1 ); the specific update process is as follows:

[0200] Calculate the advantage function A t

[0201] The advantage function is used to measure how good an action is relative to the average level; the calculation formula is:

[0202] A t =r t +γV(s t+1 |θ V )-V(s t |θ V )

[0203] Among them, A t : The advantage function represents taking an action a at time step t t The advantage of the action is a measure of how good the action is relative to the average level. The higher the advantage function value, the better the action is relative to the average level.

[0204] r t : Immediate reward, the reward obtained after executing action at at time step t; this reward reflects the direct impact of the action at the current time step on the target;

[0205] γ: discount factor, ranging from 0≤γ≤10. The discount factor is used to weigh the importance of current rewards and future rewards. A higher γ value indicates that future rewards are more important, and a lower γ value indicates that more attention is paid to current immediate rewards.

[0206] V(s t+1 |θ V )The value function of the next state represents the state s at time step +1 t+1 The expected return under the condition; the value function is calculated by the value network, and the parameter is θ V ; it estimates that in state s t+1 Next, the cumulative reward from this state to all future time steps;

[0207] V(s t |θ V ): The value function of the current state represents the state s at time step t t The expected return under the condition; the value function is calculated by the value network, and the parameter is θ V ; it estimates that in state s t Next, the cumulative reward from this state to all future time steps;

[0208] Update Policy Network

[0209] The strategy network is updated by maximizing the clipping objective function to ensure the stability of the strategy update; the specific formula is:

[0210]

[0211] in,

[0212] i: count used to sum all samples;

[0213] L π: The policy loss function is used to evaluate the performance of the current policy π and optimize the parameters of the policy network by maximizing the function;

[0214] π(a t |s t |θ π ): Current strategy, indicating the network parameters θ in the current strategy π 下 , in state s t Select action a t probability;

[0215] Old strategy, indicating the network parameters in the old strategy Next, in state s t Select action a t probability;

[0216] A t : Advantage function, measuring the t Next select action a t The relative advantages and disadvantages of t =r t +γV(s t+1 |θ V )-V(s t |θ V );

[0217] Clipping operation limits the scope of policy updates to prevent the new policy from deviating too far from the old policy;

[0218] ε: Clipping coefficient, the amplitude of the control strategy update, usually a small value;

[0219] Update value network

[0220] The value network is updated by minimizing the value function error so that it can more accurately estimate the value of each state; the specific formula is:

[0221]

[0222] L V : Value function loss, which is used to measure the error between the value network prediction value and the target value, and optimize the parameters of the value network by minimizing the loss;

[0223] By minimizing this error, the value network is able to better estimate the long-term reward of each state.

[0224] Numerical case study

[0225] The following feed samples and corresponding animal growth data were collected:

[0226]

[0227] Data preprocessing

[0228] Data cleaning

[0229] Check for missing values:

[0230] Check each sample to identify if there are any missing values. For example, the EE value of sample 2 is missing and this missing value needs to be handled.

[0231] Handling missing values:

[0232] Using mean imputation to fill missing values ​​preserves the overall trend and distribution of the data. For example, for EE, mean imputation

[0233]

[0234] Deleting samples with missing values ​​can avoid the impact of incomplete data on the analysis results, but it will reduce the amount of data.

[0235] Check for outliers:

[0236] Use the z-score method to identify outliers. Assume that the Weight of sample 3 is an extreme value, and calculate it through the z-score

[0237]

[0238] If z exceeds 3 standard deviations, it is considered an outlier.

[0239] Handling outliers:

[0240] Use the median replacement method to handle outliers. Assuming that the Weight of sample 3 is identified as an outlier, then:

[0241] Weight new =median(Weight 1 ,Weight 2 )=(55,58)=56.5

[0242] Standardization

[0243] Each nutrient component data is standardized so that its mean is 0 and its standard deviation is 1. For example, for the standardized processing of CP:

[0244]

[0245] Assuming the mean μ = 12.33 and the standard deviation σ = 1.53, the CP standardization of sample 1 is:

[0246]

[0247] Feature Engineering

[0248] Polynomial Features:

[0249] Generate polynomial features between the components to represent the interaction between the two components. For example, the polynomial feature of sample 1: NewFeature = CP × EE = 12 × 6 = 72 interaction features:

[0250] Generate CP / EE ratio features. For example, the interaction feature of sample 1:

[0251] Time series characteristics:

[0252] Consider the features of time series data. For example, for sample 1, if the data of the first two feedings is NewFeature t =f(CP t-1 ,CP t-2 )=mean(12,11)=11.5

[0253] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A method for quickly evaluating feed potency, It is characterized in that The following steps are involved: The feed to be measured is placed on a weighing machine, and based on the weighing result, the weight of the feed to be measured is determined, and the operation is repeated to quantitatively distribute the feed sample into multiple portions; The feed to be tested is sampled, crushed, sieved, floated, decolorized, digested and dissociated, and mounted on a slide, and then observed under a stereo microscope to determine the nutritional components of the feed to be tested; Based on the determined nutritional components of the feed to be tested, combined with the measured weight of the feed, the nutritional components in the feed are quantitatively analyzed, and the nutritional components include crude protein CP, ether extract EE, crude fiber CF, ash Ash, water DM, nitrogen-free extract NFE and trace elements, and the trace elements include calcium, phosphorus, salt, and vitamins; According to the results of feed nutrient analysis, the content and proportion of various nutrients in the feed are evaluated. Moisture DM content: Put the quantitative feed sample into the moisture detector to measure the moisture and volatile matter to obtain the moisture and volatile matter content; Ash content: Put the quantitative feed sample into the burner and burn it at 500-600℃ for 2h to get the mineral content; Mineral energy (%) = (mineral content / daily recommended intake) × 100%; Crude protein CP content: Put the feed sample into a container, digest the feed sample with concentrated sulfuric acid to convert all organic nitrogen into inorganic ammonium salts, then convert the ammonium salts into ammonia under alkaline conditions, distill it out with water vapor and absorb it with excess boric acid solution, then titrate it with standard hydrochloric acid to calculate the nitrogen content in the sample. The nitrogen content in protein usually accounts for 16% of its total mass. Therefore, by measuring the nitrogen content in the substance, the total protein content in the substance can be estimated, that is: protein content = nitrogen content / 16%; protein energy (%) = (protein content × 4 kcal / g) / total energy × 100%; Ether extract EE content: Mix the feed sample with ether to dissolve the fat in the ether, then separate the solution from the residue by filtering or centrifuging, and finally evaporate the ether by heating. The remaining residue is the fat content of the feed sample; Fat energy (%) = (fat content × 9 kcal / g) / total energy × 100%; Crude fiber CF content: Put the feed sample into weak acid and weak alkali solution and boil for 30 minutes. Use weak acid solution and weak alkali solution to destroy or dissolve some non-cellulose substances that are closely bound to cellulose. The boiling process helps to accelerate the chemical reaction and make the treatment more thorough. Then the treated solution is separated into solid and liquid to obtain the crude fiber residue after acid and alkali treatment, that is, the crude fiber content. The calculation formula is: crude fiber content (%) = (burning residue mass / feed sample mass) × 100%; crude fiber CF energy (%) = (crude fiber CF content / daily recommended intake) × 100%; NFE content of nitrogen-free extract: NFE% = 100% - (DM% + CP% + EE% + CF% + Ash%); Carbohydrate energy (%) = (carbohydrate content × 4 kcal / g) / total energy × 100% Calcium content: The organic matter in the feed sample is destroyed, and the calcium is converted into water-soluble ions, which are quantitatively precipitated with ammonium oxalate, and the calcium content is indirectly determined by the potassium permanganate method; Phosphorus content: First, the organic matter in the feed sample is destroyed to release phosphorus, and then it is treated with ammonium vanadium molybdate in an acidic solution to produce yellow (NH4+)3PONH4+VO3·16MoO3. The colorimetric determination is performed at a wavelength of 420nm to measure the total phosphorus content, which includes phytic phosphorus that is difficult for animals to absorb. Salt content: Under acidic conditions, add excess silver nitrate solution to make the chloride in the sample form silver chloride precipitation. After removing the precipitation, use ammonium thiocyanate to back-titrate the excess silver nitrate. Calculate the chloride content based on the amount of ammonium thiocyanate consumed. Vitamin content: Heat and saponify the feed sample with potassium hydroxide-ethanol solution to release the vitamins, then extract with organic solvents and analyze and determine with high performance liquid chromatography to achieve qualitative and quantitative analysis of multiple vitamins in the feed. During the sample processing, the antioxidant 2,6-di-tert-butyltoluene (BHT) needs to be added for protection; According to the results of feed component analysis, the content and proportion of various nutrients in the feed are evaluated. According to the nutritional requirements of different animals, it is analyzed whether the feed meets the needs of animal growth and production, so as to evaluate the potency of the feed; The absolute potency and relative potency of feed are also important indicators for evaluating feed potency. The absolute potency reflects the content of active ingredients in the feed, while the relative potency compares the difference between the feed sample and the standard feed sample. Based on the content and energy of various nutrients in the feed, calculate the ratio of animal weight gain to feed consumption and evaluate the feed conversion rate; Based on the content and energy of various nutrients in the feed, the digestibility of each nutrient in the feed is determined by collecting and analyzing animal feces.

2. A method for rapidly evaluating feed potency according to claim 1, Features: The method for calculating the ratio of animal weight gain to feed consumption and evaluating feed conversion rate is as follows: Record the initial body weight W1 of the animal and the final body weight W2 after feeding for a period of time, and calculate the weight gain G of the animal: G = W1-W2; Record the total amount of feed consumed by the animal during the specified time period F, in kilograms; Calculate the feed conversion rate CR: CR = G / F; Among them, CR is the feed conversion rate, which indicates the animal weight gain that can be achieved per kilogram of feed consumed.

3. A method for rapidly evaluating feed potency according to claim 1, Features: The method for collecting and analyzing animal feces is: The pre-feeding period for metabolic testing of poultry should be no less than 4 days, and feces should be collected continuously for 3 days afterwards; Cages should be used for raising poultry, and clean, easy-to-collect feces collection devices, including feces collection trays, plastic films or kraft paper, should be placed under the cages to accurately and completely collect poultry excrement. The collected feces samples need to be properly processed, and impurities such as feed, feathers and dandruff that contaminate the samples should be picked out. The collected feces samples should be dried, ground, sieved and stored; Chemical analysis methods, including spectral analysis and chromatographic analysis, are used to determine the content of various nutrients in fecal samples. These nutrients include protein, fat, cellulose, minerals and vitamins. The digestibility of the feed is calculated by comparing the content of nutrients in the feed and feces.

4. A method for rapidly evaluating feed potency according to claim 3, Features: The method for determining the content of various nutrients in the feces sample and evaluating the real available energy of the feed in the animal body is: Digestible energy DE is divided into apparent digestible energy ADE and true digestible energy TDE; ADE refers to the total energy of feed minus the energy in feces, that is, the undigested energy in feed, including the excretions and secretions from the animal's digestive tract, and the endogenous energy produced by intestinal microorganisms; TDE is the apparent metabolizable energy plus the endogenous energy value normally produced in the animal body; Metabolizable energy ME refers to the energy of feed nutrients that participate in the transformation of animal bodies, that is, the remaining energy after deducting fecal energy, urine energy and methane energy from the total energy value of feed. ME is divided into apparent metabolizable energy AME and true metabolizable energy TME. TME is AME plus endogenous energy in urine and feces. In order to compare the metabolizable energy values ​​of different feeds, the influence of nitrogen deposition on the ME value should be eliminated, that is, the metabolizable energy should be corrected according to the nitrogen deposition to make it the ME when the nitrogen deposition is zero. The correction formula of apparent metabolizable energy AME and true metabolizable energy TME is: AMEn=AME-RN×34.39 TMEn=TME-RN×34.39 Where: RN is the amount of nitrogen deposited by poultry daily, which can be positive, negative or zero. The sign is substituted in the calculation. 34.39 is the energy corresponding to each gram of urinary nitrogen; TME stands for true metabolizable energy, which is the energy actually absorbed and utilized by the animal body, that is, the energy after deducting fecal energy, urine energy and endogenous nitrogen loss in the digestive tract. Its calculation formula is generally: TME = TDE-[(UE-UeE)+Eg]; where TDE is total energy digestible, UE is urine energy, UeE is endogenous urine energy, and Eg is energy loss of endogenous nitrogen in the digestive tract; The relationship between metabolizable energy, digestible energy and crude protein is: ME=DE×[(96-0.202×CP) / 100] That is, for every 1 percentage point increase in crude protein, the utilization rate of digestible energy converted into metabolic energy decreases by 0.202 percentage points.

5. A method for rapidly evaluating feed potency according to claim 1, Features: Based on the determined nutritional components of the feed to be measured, combined with the measured weight of the feed, the nutritional components in the feed are quantitatively analyzed. The specific process is as follows: Introducing a model-based reinforcement learning intelligent algorithm to quantitatively analyze and optimize the nutritional components in feed; Step 1: Data collection and preprocessing Data collection: Collect nutritional data of a large number of feed samples and animal growth data for subsequent analysis and modeling; specifically includes the following components: Crude protein CP: Protein is the main nutrient required for animal growth and repair of body tissues; crude protein is the sum of all protein components in feed, including protein from animal and plant sources; Ether extract EE: refers to the total amount of fat and oil in feed extracted by ether; these ingredients provide high energy for animals and help absorb fat-soluble vitamins, vitamins A, D, E and K; the appropriateness of fat content directly affects the energy intake and balance of animals; Crude fiber CF: refers to the cellulose part of feed that is insoluble in acid and alkali, mainly including cellulose, hemicellulose and lignin; Ash: refers to the total amount of minerals in feed, including calcium, phosphorus and magnesium; Moisture DM: refers to the moisture content in feed; moisture content affects the preservation and use of feed. Too much moisture causes the feed to mold, while too little moisture affects the palatability and digestibility of the feed. Nitrogen-free extract (NFE): refers to carbohydrates, starches and sugars in feed that do not contain nitrogen; these ingredients provide energy for animals and are an important source of energy for animal activity and growth; Trace elements: including calcium, phosphorus, salt, vitamins, these elements in small amounts are essential for the growth, development and health of animals; although trace elements and vitamins are required in small quantities, their deficiency can cause serious health problems; Data preprocessing: Data cleaning: Check missing values: For each sample, check whether there are missing values; missing values ​​are incomplete data due to measurement errors or recording errors; Identify these missing values ​​to ensure data completeness and accuracy; Handling missing values: Missing values ​​can be filled with the mean to maintain the overall trend and distribution of the data; Mean filling formula: Among them, n new : The number of new data; x i is a non-missing value, and n is the number of non-missing values; deleting samples with missing values ​​can avoid the impact of incomplete data on the analysis results, but it will reduce the amount of data; Check outliers: Use the z-score method to identify values ​​far from the mean. Outliers are extreme values ​​caused by input errors or equipment failures. Values ​​exceeding 3 standard deviations are usually considered outliers. The z-score formula is: Among them, z is the recognition criterion, x is the sample value, μ is the mean, and σ is the standard deviation; Handling outliers: Use the median replacement method to handle outliers. Median replacement can reduce the impact of outliers on data distribution, thereby improving the reliability of analysis results. The median replacement method effectively reduces the impact of outliers on data because the median is less affected by extreme values. Median replacement formula: x new =median(x 1 ,x 2 ,...,x n ) x new : New data value, that is, the data value after replacing abnormal values ​​or missing values; (x 1 ,x 2 ,...,x n ): data values ​​in the original data set, which do not contain outliers or missing values; that is, only normal, reasonable data points are used to calculate the median; median(x 1 ,x 2 ,...,x n ): The median of the original data set; the median is the value in the middle after all the data are arranged in order of size; for an odd number of data points, the median is the middle number; for an even number of data points, the median is the average of the two middle numbers; Standardization: Each nutrient component data is standardized so that its mean is 0 and its standard deviation is 1. The purpose of standardization is to convert data of different dimensions to the same scale so that each component contributes equally to the model. This helps speed up the model training process and improves the performance of the model. After standardization, the distribution of each component data is more consistent, which is conducive to the rapid convergence and stability of the algorithm. Among them, x` is the numerical value corresponding to each data value after standardization, x is the sample value, μ is the mean, and σ is the standard deviation; Feature extraction: Introducing feature extraction to enhance the expressiveness of data by combining, selecting or generating new features; feature engineering can mine more useful information and improve the predictive performance of the model; Polynomial features: Generate polynomial features between the components; Generate the product feature of CP and EE to represent the interaction between the two components; By generating polynomial features, the nonlinear relationship between feed components is captured; Official: NewFeature=q 1 ×q 2 NewFeature: new feature; q i : An original characteristic, any two of crude protein CP, ether extract EE, crude fiber CF, ash Ash, moisture DM, nitrogen-free extract NFE, and trace elements; Interaction features: Combine the interaction features between components; generate CP / EE ratio features to represent the relative relationship between the two components; reveal the relative importance and mutual influence of feed components; Time series characteristics: Consider the characteristics of time series data, the relationship between the ingredients of the previous feedings and the current ingredients; this can help the model capture the cumulative impact of feed ingredients on animal growth; reflect the changing trend of feed ingredients over time and the long-term impact on animal growth; formula: NewFeature t =f(NewFeature t-1 ,NewFeature t-2 ,...,NewFeature t-n ) Among them, NewFeature t : The new feature at time t; this is the feature value calculated by function f based on the feature values ​​of the previous n time steps; f: Function f represents the mathematical relationship or algorithm for converting the feature values ​​of the previous n time steps into the new feature value of the current time step. This function can be in the form of weighted average, cumulative sum, or difference function; NewFeature t-n : The eigenvalue at time tn; represents the eigenvalue at the nth time step before the current time step; Step 2: Environmental model construction State representation: define state s as the nutritional composition of the current feed formula and the growth status of the animal; the state vector contains the specific values ​​of different components and the weight information of the animal; State vector: s t =[CP,EE,CF,Ash,DM,NFE,Ca,P,Salt,Vitamin,Weight] Specific explanation: Ca: calcium, an important mineral element; P: phosphorus, an important mineral element; Salt: salt, an important component for maintaining electrolyte balance; Vitamin: vitamin, a micronutrient that supports growth and immune function; Weight: animal weight, indicating the current weight of the animal; Action representation: define action a t To make decisions on adjusting feed formulation, it means increasing or decreasing the proportion of a certain ingredient; Action Vector: a=[ΔCP,ΔEE,ΔCF,ΔAsh,ΔDM,ΔNFE,ΔCa,ΔP,ΔSalt,ΔVitamin]t Specific explanation: ΔCP: adjustment of crude protein; ΔEE: adjustment of ether extract; ΔCF: adjustment of crude fiber; ΔAsh: adjustment of ash; ΔDM: adjustment of water; ΔNFE: adjustment of nitrogen-free extract; ΔCa: adjustment of calcium; ΔP: adjustment of phosphorus; ΔSalt: adjustment of salt; ΔVitamin: adjustment of vitamins; State transfer model: It uses a graph neural network-based approach to consider the relationships and dependencies between components. The graph neural network model can effectively capture complex relationships and nonlinear effects, providing more accurate predictions. State transition formula s t+1 =GNN(s t ,a t ) Specific explanation: st+1: represents the new state at time t+1, that is, the changes in feed composition and animal weight after executing action at; The input of the graph neural network model is the current state st and action at, and the output is the new state st+1; GNN; graph neural network function; Input layer: The current state s and action a are used as input; each input node represents a state vector tt Quantitative ingredients; Graph construction: Build a graph structure containing all components and animal weights, and the relationships between components are connected by edges; Feature extraction: The features of each node are extracted through a multi-layer graph neural network model layer; the graph neural network model layer aggregates the information of each node with the information of its adjacent nodes through neighborhood aggregation operations to form a new feature representation; Output layer: Outputs the new state st+1, which is the change of each component and the weight of the animal after the action is performed; Mathematical expression of graph neural network model: Each layer of the graph neural network model updates its features using the following formula: Where: v: node; u: neighbor node; The initial features represent the features of node v in the kth layer; W(k) represents the characteristics of the neighbor node u in the kth layer; N(v) represents the set of neighbor nodes of node v; b(k) The weight matrix of the kth layer; the information of neighbor nodes is linearly transformed and then aggregated; the bias vector of the kth layer; the bias term is introduced to improve the expressiveness of the model; this vector is also a parameter that needs to be learned through training; σ is a nonlinear activation function; the introduction of nonlinearity enables the model to capture complex relationships; is the feature representation of node v in the k+1th layer; Finally, after passing through several layers of graph neural network models, a new feature representation of each node is obtained, namely s New state t+1; Step 3: Reward Function Design The reward function r is the key to evaluating the effect of an action in a specific state in reinforcement learning. Components; In the animal feed formulation optimization problem, the design of the reward function should reflect the animal's weight gain effect and feed utilization efficiency; an effective reward function can guide the model to make optimization decisions, improve the feed's potency and the animal's health; The reward function can be defined as the ratio of animal weight gain to feed consumption Specific explanation: r: reward value at time t, reflecting animal weight gain and feed utilization efficiency; t ΔWeightt: animal weight gain at time t, indicating the change in animal weight;FeedConsumptiontt: feed consumption at time t; Step 4: Reinforcement Learning Strategy Optimization Strategy representation: define strategy π(at|st) as the probability of selecting action at in state st; the strategy network is used to approximate this strategy, input the current state s, and output the probability distribution of different actions a; this means This means that in a given state, the policy network will calculate the probability of selecting each possible action, thereby guiding the decision-making process; Algorithm selection: The model-based reinforcement learning algorithm PPO is used. The specific process is as follows: initialization Policy network π(st|θπ): used to output the probability distribution of each action in each state; here θπ represents the parameters of the policy network; these parameters will be optimized through training so that the policy network can correctly estimate the probability of the best action; Value network V(st|θV): used to estimate the value function of each state, that is, the expected return in state st; here θV represents the parameters of the value network; the value network helps the policy network evaluate the potential value of each state; Experience replay buffer R: used to store the experience data of each step, including state, action, reward and next state; these experience data will be used to update network parameters in subsequent training; Action Selection At each time step, an action at = π(st|θπ) is selected from the policy network; the specific process is as follows: Input the current state s into the policy network and output the probability distribution of each action; t Sampling is done based on the output probability distribution, and action a is selected; this means that if the t of an action The higher the probability, the higher the chance that it will be selected; Storage experience The experience (st,at,rt,st +1 ) is stored in the experience replay buffer R; these experience data are used for subsequent network updates; Update Network Randomly sample mini-batches (st, at, rt, st) from the experience replay buffer +1 ); the specific update process is as follows: Calculate the advantage function A t The advantage function is used to measure how good an action is relative to the average level; the calculation formula is: At=rt+γV(st +1 |θV)-V(st|θV) Among them, A: advantage function represents the advantage of taking an action a at time step t, measuring the action tt The value function of the current state represents the expected return in state st at time step t. The value function is calculated by the value network with parameter θV. It estimates the expected return from starting from state s in state s. The cumulative reward from the beginning to all future time steps; Update Policy Network The strategy network is updated by maximizing the clipping objective function to ensure the stability of the strategy update; the specific formula is: Where i: count for summing all samples; L: policy loss function used to evaluate the current policy The performance of the policy network is slightly smaller than π, and the parameters of the policy network are optimized by maximizing this function; N: the number of samples; π(at|st|θπ): current strategy, indicating that action t is selected in state s under the current strategy network parameters θπ The probability of a; π(at|st|θolπd): old strategy, which means the probability of selecting action a in state stt under the old strategy network parameters θoπld; A: advantage function, which measures the relative merits of selecting action a in state s; tttt The calculation formula is: At = rt + γV (st +1 |θ V )-V(st|θV); clip operation limits the amplitude of policy update to prevent the new policy from deviating too far from the old policy; ε is the clip coefficient, which controls the amplitude of policy update and is usually small; Update value network The value network is updated by minimizing the value function error so that it can more accurately estimate the value of each state; the specific formula is: LV: value function loss, which is used to measure the error between the value network prediction value and the target value, and optimize the parameters of the value network by minimizing the loss; V(st +1 |θV) represents the expected reward at state st+1 at time step +1; the value function is calculated by the value network with parameter θV; it estimates the cumulative reward at state st+1, starting from that state to all future time steps; by minimizing this error, the value network is able to better evaluate the long-term reward of each state.

Citation Information

Cited By

  • Animal feeding plan generation method and system based on ingestion monitoring data

    CN119378956A