Search

CN-122021339-A - Intelligent optimization method for low-phosphorus ration formula of laying hen based on Bayesian optimization and neural network

CN122021339ACN 122021339 ACN122021339 ACN 122021339ACN-122021339-A

Abstract

A low-phosphorus ration formula intelligent optimization method for laying hens based on Bayesian optimization and a neural network is characterized by comprising the following steps of S1, constructing an integrated data set for model training, S2, constructing a weighted multi-target neural network prediction model, and S3, executing Bayesian optimization based on a fusion agent model. The invention radically innovates the research and development mode of the low-phosphorus ration formula of the laying hen, and realizes the synergistic maximization of production performance, product quality, economic benefit and environmental benefit.

Inventors

  • ZHANG YU
  • LI XIANLEI
  • ZHAO XIANGHONG
  • ZHANG BINGKUN
  • ZHANG YAQIAN

Assignees

  • 北京中农优嘉生物科技有限公司

Dates

Publication Date
20260512
Application Date
20260226

Claims (10)

  1. 1. A low-phosphorus ration formula intelligent optimization method for laying hens based on Bayesian optimization and neural network is characterized by comprising the following steps: S1, constructing an integrated data set for model training: S11, acquiring first basic data, namely acquiring a plurality of first data points from a low-phosphorus feed test of the laying hens, wherein each first data point comprises a calcium hydrogen content CP and a selection identifier of a first or second type phytase Amount of phytase added And the corresponding laying rate Thickness of eggshell He Ha's unit ; S12, obtaining second reference data, namely obtaining reference data from a phosphorus metabolism test of broiler chickens, wherein the reference data comprise apparent digestibility of basic phosphorus of a phytase-free control group Phosphorus apparent digestibility enhancement value characterizing the efficacy of the first type and the second type phytase respectively Wherein ; S13, constructing a cross-species mapping relation, namely, based on a common physiological foundation formed by animal phosphorus metabolism and eggshells, utilizing a concept of phosphorus digestibility corresponding to eggshell thickness in the first basic data and second basic data, and establishing a predicted value from phosphorus apparent digestibility D to eggshell thickness through regression analysis Is a mapping function of: ; S14, generating a key virtual control point, and enabling the apparent digestibility of the basic phosphorus to be improved Substitution into the mapping function Calculating to obtain a virtual eggshell thickness value And calculating the corresponding virtual laying rate according to the laying law of the laying hens And virtual Hastell units Thereby generating a virtual control data point characterizing the conditions of zero calcium hydrogen and no phytase ; S15, generating a supplementary test point based on the efficacy enhancement value of the second type phytase Generating a series of supplementary data point sets containing zero calcium hydrogen conditions and different amounts of the phytase of the second type by taking the data points containing the phytase of the second type with the optimal performance in the first basic data as references Wherein Based on the value of (2) Determining a mapping relation with a reference point; S16, merging and constructing the data set, namely merging all points of the first basic data and the virtual control data point And all points in the supplemental data point set S are combined to form an integrated data set which covers from high to zero calcium hydrogen, contains two types of phytase and has key control information ; S2, constructing a weighted multi-target neural network prediction model: S21, integrating the data set Performing characteristic engineering to at least construct the following characteristics of hydrogen-calcium content CP and phytase type identification Amount of phytase added Features of interaction ; S22, constructing the feed with the characteristics as input and the laying rate Thickness of eggshell Ha's unit A feedforward neural network model for an output target; s23, training the neural network model by adopting a weighted mean square error as a loss function, wherein the loss function is that Wherein Is of preset weight and ; S3, executing Bayesian optimization based on fusion agent model: s31, defining a comprehensive benefit function Wherein For a weight coefficient greater than zero, And (3) with Costs for calcium hydride and phytase, respectively; s32, combining the neural network model and a Gaussian process model to construct a weighted fusion proxy model, wherein the weight ratio of the neural network output is greater than 0.5; s33, carrying out Bayesian optimization iteration by using the fusion agent model in a formula parameter space with the aim of maximizing the comprehensive benefit function F, and being constrained by: wherein A preset threshold value; s34, outputting optimal formula parameters when the optimization iteration converges.
  2. 2. The method according to claim 1, wherein in step S13, the mapping function As a linear function: where k and b are regression coefficients.
  3. 3. The method according to claim 1 or 2, characterized in that in step S12, the phosphorus apparent digestibility is increased by a value 7.16 Percent, 11.02%.
  4. 4. The method of claim 1, wherein in step S21, the feature engineering further comprises constructing interactive features 。
  5. 5. The method according to claim 1, wherein in step S23, the weight 0.4,0.3,0.3 Respectively.
  6. 6. The method according to claim 1, wherein in step S32, weights of the neural network model and the gaussian process model are 0.7 and 0.3, respectively.
  7. 7. The method according to claim 1, wherein in step S33, the constraint threshold is 90%,0.31Mm,82.0 respectively.
  8. 8. The method according to claim 1, wherein in step S31, the weight coefficients 0.35,0.25,0.20,0.10,0.10 Respectively.
  9. 9. The method according to any one of claims 1 to 8, wherein the optimal recipe parameters outputted in step S34 are: in the second category of the products, 。
  10. 10. A low phosphorus ration for laying hens, characterized in that the formulation is determined by the method according to any one of claims 1-9, and the formulation comprises 0% calcium hydride and 120g/t phytase of the second type.

Description

Intelligent optimization method for low-phosphorus ration formula of laying hen based on Bayesian optimization and neural network Technical Field The invention relates to the technical field of livestock and poultry feed formula optimization and artificial intelligence intersection, in particular to an intelligent optimization method for a low-phosphorus ration formula of a laying hen based on Bayesian optimization and a neural network. Background The laying hen breeding industry is an important component part of the livestock industry in China, and the annual feed consumption is huge. Phosphorus is a key mineral element for bone development and eggshell formation of laying hens, and in order to ensure sufficient supply of phosphorus, inorganic phosphorus sources such as calcium hydrophosphate (hydrogen calcium) and the like with the concentration of about 0.4% are usually added in the conventional daily ration. This not only increases the feed cost by about 3-5%, but also causes serious environmental problems. According to statistics, about 1.8g of phosphorus is discharged per 1 kg of eggs, and the annual phosphorus discharge amount is more than 10 ten thousand tons, so that the method is one of important pollution sources for water eutrophication. The phytase can effectively decompose the phytate phosphorus in the feed and release available phosphorus, thereby providing possibility for reducing or even replacing inorganic phosphorus addition. The prior art has explored this, for example, the chinese patent literature with publication number CN108208346a discloses "a low-phosphorus complete ration for laying hens in peak laying period and application thereof". The technical scheme proposes feeding in two stages, namely, using daily ration added with 0.02% phytase and without adding exogenous phosphorus source (such as calcium hydrogen) in 20-40 weeks, and using daily ration without adding phytase and exogenous phosphorus source in 41-55 weeks, and claiming that the nutritional requirements of laying hens can be met, saving about 17.6 kg/ton of feed of calcium hydrogen phosphate and reducing phosphorus emission. Although the prior art to a certain extent verifies the feasibility of completely canceling the addition of inorganic phosphorus at a specific stage, the technical scheme and the similar researches still have the following obvious limitations, and the challenges facing the optimization of the low-phosphorus ration formula cannot be systematically solved: First, the recipe determination method is inefficient, costly and of limited accuracy. Existing studies, represented by CN108208346a, typically employ a traditional "one-way gradient test design", i.e., a long-term feeding test with multiple fixed calcium hydrogen addition levels (e.g., 0%, 0.05%, 0.10%. 0.42% gradient) to "screen" for a viable low limit by comparing production performance for each group. This method has a long experimental period (e.g., 31 weeks of testing in the examples), requires a large number of animals, and is extremely costly. More importantly, the method is essentially a discretized 'grid search', can only verify preset points, cannot accurately position the optimal solution (such as the most accurate addition of phytase) in a continuous formula parameter space, and is difficult to describe complex nonlinear interactions among factors. Second, existing data and knowledge are not fully utilized. Existing optimization methods tend to be limited to the data generated by the current single trial, e.g., the scheme of CN108208346a relies entirely on two complete sets of gradient trials performed separately for "dibasic calcium phosphate" and "dibasic calcium phosphate" in examples 1 and 2. The method does not fully utilize historical research data (such as a large amount of application data about different phytase formulations and activities on broiler chickens or other poultry species), and each optimization is almost started from zero, so that huge scientific research resource waste and data value loss are caused. Third, there is a lack of systematic trade-offs for multi-objective collaborative optimization. The prior art has focused on whether a single or few core objectives of "maintenance of productivity" are achieved (e.g., egg laying rate > 85%). However, in practical production, formula optimization needs to achieve an optimal balance between multiple interrelated and even conflicting objectives such as production performance (egg yield), product quality (eggshell thickness, hawster's unit), economic cost (calcium hydride, phytase price), and environmental benefit (phosphorus emission reduction). The solution of CN108208346a gives a static, fixed formulation (e.g., 0.02% phytase), but does not provide a method that dynamically balances cost and benefit, rapidly responds to raw material price fluctuations, and accurately quantifies the comprehensive optimal solution. Fourth, the universality and mobility of the solution ar