CN-122020174-A - Method and system for collaborative prediction of growth performance of live pigs and environmental excrement

CN122020174ACN 122020174 ACN122020174 ACN 122020174ACN-122020174-A

Abstract

The invention discloses a method and a system for collaborative prediction of growth performance of live pigs and environmental excreta, and belongs to the technical field of intelligent agriculture and artificial intelligence. The method comprises the steps of constructing a layered hybrid prediction framework by fusing a Bayesian mechanism model and a machine learning residual error correction model, firstly establishing the Bayesian mechanism model based on multi-source heterogeneous data to generate a preliminary prediction result with biological interpretability, then taking the mechanism model output residual error as a learning target, constructing the machine learning residual error correction model to refine the preliminary prediction result, and finally revealing residual error correction driving factors through an interpretability analysis tool to form a collaborative prediction report of growth performance and nitrogen and phosphorus excretion. The method and the device remarkably improve the prediction precision while maintaining the interpretability of the mechanism model, realize unified dynamic prediction of daily gain, feed efficiency and environmental excrement of the live pigs, and provide decision support for precise feeding and environment-friendly cultivation.

Inventors

JIANG HONGHUA
HOU TIANQI
YANG JIAJUN
JIANG SHUZHEN
SHI SHAOYANG
YANG WEIREN

Assignees

山东农业大学

Dates

Publication Date: 20260512
Application Date: 20260204

Claims (10)

1. A method for collaborative prediction of growth performance and environmental excreta of live pigs is characterized by comprising the following steps: acquiring multisource data of live pigs, and preprocessing the multisource data to obtain standardized input data; Constructing a first-stage Bayesian mechanism model according to the standardized input data, and generating a preliminary prediction result; obtaining residual errors according to the difference between the preliminary prediction result and the actual observation value; Constructing a second-stage machine learning residual correction model according to the residual, and generating a residual predicted value; Obtaining a final prediction result according to the preliminary prediction result and the residual error prediction value; and outputting a collaborative prediction report of the growth performance and the environmental excreta according to the final prediction result.
2. The method of claim 1, wherein the obtaining standardized input data comprises: Acquiring a core data set according to animal performance data and nutrition intake data; Acquiring a formula data set according to feed raw material composition data; and integrating data according to the core data set and the formula data set to obtain the standardized input data.
3. The method of claim 2, wherein the acquiring the core dataset comprises: acquiring a performance subset according to initial weight, test days and daily gain data; Acquiring a nutrition subset according to intake data of crude protein, net energy, amino acid, total phosphorus, digestible phosphorus and calcium; the core dataset is constructed from the performance subset and the nutritional subset.
4. The method of claim 1, wherein the generating preliminary prediction results comprises: setting prior distribution of model parameters according to the standardized input data; according to the prior distribution, deducing model parameters through a Markov chain Monte Carlo sampling algorithm, and calculating intermediate metabolic variables of a nutrient substance metabolic path; and calculating protein deposition, fat deposition, nitrogen excretion and phosphorus excretion according to the intermediate metabolic variables, and generating the preliminary prediction result.
5. The method of claim 4, wherein the obtaining an intermediate metabolic variable comprises: Constructing a genetic maximum protein deposition potential curve, and simulating the descending trend of the protein deposition capacity of the live pigs in the later growth period by adopting a double-logic Studies difference function; Calculating a plurality of potential protein deposition amounts limited by different amino acids, and weighting and integrating genetic potential and amino acid limiting potential by applying a differentiable smooth minimum function to obtain effective protein deposition potential; The growth requirement of phosphorus is decomposed into soft tissue phosphorus requirement related to protein deposition and bone phosphorus requirement related to ash deposition, and the soft tissue phosphorus requirement and the bone phosphorus requirement are calculated and added independently based on the protein deposition amount and the ash deposition amount respectively to obtain the total phosphorus growth requirement.
6. The method of claim 1, wherein the obtaining the residual comprises: calculating a residual value according to the actual daily gain data and the daily gain in the preliminary prediction result; and constructing a residual data set according to the residual values, and training a second-level model.
7. The method of claim 1, wherein generating the residual prediction value comprises: Constructing a feature set according to the residual error; And training a gradient lifting decision tree model according to the feature set to obtain the residual prediction value.
8. The method of claim 7, wherein constructing the feature set comprises: acquiring raw material polymerization characteristics according to functional classification of the feed raw materials; acquiring nutrition interaction characteristics according to the nutrient proportion relation; acquiring mechanism linkage characteristics according to intermediate metabolic variables in the preliminary prediction result, wherein the mechanism linkage characteristics comprise: a digestible phosphorus balance for characterizing the nonlinear effect of the degree of deviation of phosphorus supply from biological demand on growth performance; producing a net energy duty cycle for characterizing a competitive pressure required to sustain energy distribution; a restriction factor index for conveying to the second level model a specific nutrient class currently restricting growth, the auxiliary model identifying a growth bottleneck; and constructing the feature set according to the raw material polymerization feature, the nutrition interaction feature and the mechanism linkage feature.
9. The method of claim 1, wherein the obtaining the final prediction result comprises: Calculating the final daily gain according to the daily gain in the preliminary prediction result and the residual prediction value; Based on the mass balance principle, the total nitrogen excretion and the total phosphorus excretion are calculated respectively: total nitrogen excretion = total nitrogen intake-model predicted nitrogen deposition; total phosphorus excretion = total phosphorus intake-model predicted phosphorus deposition; wherein the phosphorus excretion is further divided into: Fecal phosphorus excretion = total phosphorus intake-standard ileal digestible phosphorus intake; urinary phosphorus excretion = standard ileal digestible phosphorus intake-model predicted phosphorus deposition; and constructing the final prediction result according to the final daily gain and the nitrogen and phosphorus excretion in the preliminary prediction result.
10. A live pig growth performance and environmental waste collaborative prediction system, configured to perform a method according to any one of claims 1 to 9.

Description

Method and system for collaborative prediction of growth performance of live pigs and environmental excrement Technical Field The invention belongs to the technical field of intelligent agriculture and artificial intelligence, and particularly relates to a method and a system for collaborative prediction of growth performance of live pigs and environmental excreta. Background In the modern intensive live pig breeding industry, realizing accurate feeding is a core link of improving production efficiency, reducing feed cost and reducing environmental pollution. The key point is to dynamically and accurately predict the future growth performance (such as daily gain and feed efficiency) of pigs under specific daily ration, genetic background and breeding environment, and the critical environmental excretions such as nitrogen, phosphorus and the like closely related to the pig. Currently, the mainstream prediction models can be mainly divided into two main categories according to the construction principle. The first is a mechanism model, such as classical models of NRC, inrapac, etc. The model is based on the first principle of animal physiology, biochemistry and nutrition, describes the metabolic path of nutrient substances through a mathematical equation, has a clear structure, is known as a white box or a glass box, has good interpretability, and can help a nutrient to understand an internal regulation mechanism. However, such models have inherent technical drawbacks in that the core physiological metabolic parameters (such as maintenance requirements, deposition efficiency coefficients, etc.) in the model are usually set to fixed values or simple functions, and such "parameter rigidity" causes difficulty in adapting the model to dynamic changes caused by different genetic lines, health conditions and complex environmental stresses, so that systematic prediction bias is generated, and the prediction accuracy is difficult to meet the requirements of commercial accurate feeding on individual or small group level fine management. In order to compensate for the problem of insufficient accuracy of the mechanism model, a second class of data-driven machine learning models, such as gradient lifting decision trees, neural networks, and the like, are applied to the field. Such models enable learning complex nonlinear relationships between inputs and outputs from massive historical production data, often enabling higher prediction accuracy on specific data sets. However, pure machine learning models have a deadly "black box" nature, their decision process is highly complex and opaque, and cannot provide an explanation that is in line with biological logic, making it difficult for production managers to trust their predicted results, and to trace their biological sources when predictions deviate. In addition, the effect is highly dependent on the quantity and quality of training data, and for new situations where the data is uncovered (such as novel feed materials), the generalization ability and reliability of the data are at significant risk. Attempts have been made in the prior art to construct hybrid models to combine the advantages of both. However, most existing hybrid models employ either a 'parallel' structure (i.e., training a mechanism model and a machine learning model, respectively, and then weighting the results to average) or a simple 'series' structure (taking only the final output value of the mechanism model as one feature of machine learning). These methods fail to mine the 'process variables' (Process Variables) inside the mechanism model. In fact, hidden state variables such as 'phosphorus equilibrium state', 'restricted amino acid type', etc., calculated inside the mechanism model, contain extremely high value biological diagnostic information. The prior art fails to establish a 'hidden state transmission channel' from the inside of the mechanism model to the input of the machine learning model, so that the machine learning model still fits data blindly, and the biological reasons behind the growth deviation cannot be really understood. Therefore, there is an urgent need in the art for a new technical solution that can cooperatively implement high-precision and high-interpretability growth performance and environmental excreta prediction within a unified dynamic framework. Disclosure of Invention The invention provides a method and a system for collaborative prediction of growth performance of live pigs and environmental excreta, which are used for solving the problems in the prior art. In order to achieve the above object, the present invention provides a method for collaborative prediction of growth performance of live pigs and environmental excreta, comprising: acquiring multisource data of live pigs, and preprocessing the multisource data to obtain standardized input data; Constructing a first-stage Bayesian mechanism model according to the standardized input data, and generating a preliminary