Data and knowledge dual-driven accurate Hu sheep feeding method

Abstract

The invention relates to a data and knowledge dual-driven accurate Hu sheep feeding method. Comprising the following steps: collecting, processing and storing data, and constructing a Hu sheep feeding knowledge graph; according to knowledge in the knowledge graph, a feeding scheme of the Hu sheep in a certain stage and the feeding nutrition demand quantity of the Hu sheep in the stage are obtained by adopting a knowledge graph reasoning technology and a case reasoning method; a plurality of factors of feed are considered, the feeding economic benefit maximization problem is solved in combination with the nutrition demand quantity, and the optimal feeding scheme and feed feeding quantity are obtained. According to the method, the feeding benefit of the Hu sheep can be improved, the feed conversion ratio is reduced, and the method has a remarkable effect on reasonable utilization of feed in the growth process of the Hu sheep and refined development of animal husbandry.

Description

Technical Field

[0001] The present invention relates to the field of biological feeding, and in particular to a data and knowledge-driven precision feeding method for Hu sheep. Background Art

[0002] As a unique and excellent local sheep breed in my country, Hu sheep is mainly distributed in the Hangzhou, Jiaxing and Huzhou areas of Zhejiang Province, Suzhou City of Jiangsu Province and parts of Shanghai. It has excellent traits such as tolerance to rough feed, suitability for indoor feeding, and estrus in all seasons. It is also a first-class protected local livestock and poultry breed in my country and has been included in the "National Livestock and Poultry Genetic Resources Protection Catalogue" twice.

[0003] The growth and development of Hu sheep runs through the entire breeding process, and efficient Hu sheep precision feeding methods are an inevitable requirement for the high-quality development of the Hu sheep industry. Databases such as China National Knowledge Infrastructure (https: / / www.cnki.net / ) record many research papers on Hu sheep breeding.

[0004] At present, the formulation of the feeding plan for Hu sheep mainly relies on the experience of farmers. Some farmers and animal husbandry experts simply guide feeding by building their own knowledge graphs based on search engines such as China National Knowledge Infrastructure. Therefore, the effectiveness of data is not fully explored. There is a certain degree of blindness in the feeding behavior of farmers, and they are not good at dealing with numerical prediction problems. It is well known that unreasonable feeding of feed can lead to health problems of Hu sheep, decreased growth and production efficiency, feed waste, and other problems, and also increase cost consumption. Therefore, it is of great value to study the formulation method of Hu sheep precision feeding. Summary of the invention

[0005] The technical problem to be solved by the present invention is: in order to solve the problem of large feed consumption and high feeding cost in the existing Hu sheep feeding technology, a data and knowledge dual-driven Hu sheep precision feeding method is designed. The details are as follows:

[0006] Step S1: After data retrieval and extraction processing, a Hu sheep feed data set, a Hu sheep feeding guidance data set, and a Hu sheep feeding case data set are obtained; each data set is integrated and stored in a relational database to form a basic Hu sheep feeding knowledge graph;

[0007] Step S2: When there is a new feeding event, the knowledge graph representation learning is performed on the Hu sheep feeding knowledge graph, and then the feeding plan for the growth stage to be fed is predicted by the vector obtained by representation learning, combined with the existing information in the graph as the result of knowledge graph reasoning;

[0008] Step S3: Divide the overall case of Huyang feeding into a problem description part and a solution part;

[0009] Step S4: Filter out k groups of historical cases that are relatively close to the event to be recommended as the result of case retrieval based on the similarity between the event to be recommended and the problem description of each case in the case library;

[0010] Step S5: The k similar cases obtained in the case retrieval stage of Hu sheep feeding are used to modify the retrieved cases, and the global similarity between the event to be recommended and the similar cases is applied to the prediction of the daily required nutrition; the predicted values ​​obtained after adjustment of each similar case are summed up, and then divided by the sum of the similarities to obtain the required nutrition for the growth stage of the event to be recommended;

[0011] Step S6: using a feed feeding amount acquisition algorithm to obtain the most efficient feed formula for the existing data;

[0012] Step S7: Evaluate the feeding effect regularly and adjust and optimize the feeding plan according to the evaluation results.

[0013] As a preferred method: the collection of Hu sheep feeding cases and data sets are derived from feeding experiment record papers closely related to Hu sheep in CNKI. Hu sheep experiment record papers are searched in CNKI, and then information is extracted from the obtained papers. Cases with clear numerical values ​​of various indicators and complete recording processes are retained and stored in the case library to obtain a Hu sheep feeding case data set.

[0014] As a preferred embodiment: the Hu sheep feed dataset, the Hu sheep feeding guidance dataset, and the Hu sheep feeding case dataset are all obtained through template-based automatic information extraction and manual sorting and correction.

[0015] As a preferred embodiment: step S2 further comprises:

[0016] Step S21: performing information retrieval on the knowledge related to the growth stage of the Hu sheep to be fed, comparing the retrieved information types with the information types required for the recommendation results, and obtaining the missing information types in the knowledge graph;

[0017] Step S22: using the knowledge graph representation learning model to map entities and relationships into low-dimensional vectors, using the numerical entity and relationship vectors for inference and prediction, and obtaining the value of the missing information; wherein step S22 includes:

[0018] Step S221: According to step S1, construct a knowledge graph of Hu sheep feeding, using a knowledge reasoning model that can capture the semantic connections between entities;

[0019] Step S222: Based on the representation vectors of each entity and relationship in the graph obtained by representation learning, the growth stage to be fed is used as the head entity, and for each missing relationship, each entity in the graph is used as the tail entity to calculate the scoring function;

[0020] Step S223: when the feeding plan for Hu sheep obtained through information retrieval is incomplete, the missing information is supplemented using link prediction technology;

[0021] Step S23: combining the existing information in the atlas, and taking the predicted feeding plan and the existing feeding plan in the atlas as the recommendation result.

[0022] As a preferred embodiment, step S3 includes:

[0023] Step S31: The indicators of the problem description part include the growth stage of Hu sheep, feeding management, feed nutrition, daily nutrient requirements, metabolizable energy, dry matter, crude protein, neutral detergent fiber, acid detergent fiber, crude fat, crude fiber, crude ash, calcium and phosphorus;

[0024] Step S32: The solution part includes feeding environment requirements, feed type, feed content in the formula, and feed concentrate and coarse ratio;

[0025] Step S33: The values ​​of various indicators in the Huyang feeding case include metabolizable energy, dry matter, crude protein, neutral detergent fiber, acid detergent fiber, crude fat, crude fiber, crude ash, calcium and total phosphorus, which refer to ME, DM, CP, NDF, ADF, EE, CF, Ash, Ca and P respectively.

[0026] As a preferred embodiment, step S4 includes:

[0027] Step S41: In the attribute similarity calculation, the attributes are divided into two categories according to the data type: text attributes and numerical attributes; the text attributes include the growth stage of Hu sheep and feeding management, and the numerical attributes include feed nutrition, daily nutritional requirements, metabolizable energy, dry matter, crude protein, neutral detergent fiber, acid detergent fiber, crude fat, crude ash, calcium and total phosphorus;

[0028] Step S42: the similarity of the same numerical attribute between different cases is represented by the distance between the two values ​​after normalization;

[0029] Step S43: using the distributed low-dimensional vectors of each growth stage obtained after the knowledge graph representation learning as the numerical representation of each growth stage, and calculating the similarity between the growth stages using the cosine similarity between the vectors;

[0030] Step S44: Use the formula to calculate the global similarity of the case, and take the k historical cases with the highest global similarity and the lowest feed-to-meat ratio as the case retrieval results

[0031] As a preferred embodiment, step S6 includes:

[0032] Step S61: define problem variables, including feed usage vector, feed price vector, Hu sheep nutritional demand vector, feed nutritional content matrix and warehouse feed surplus vector, etc.;

[0033] Step S62: Calculate the total cost, i.e., the sum of the product of the amount of each feed and its corresponding price;

[0034] Step S63: determining an objective function, aiming to minimize the total cost;

[0035] Step S64: setting constraints, including satisfying the nutritional needs of the Hu sheep, ensuring that the amount of each feed is less than the corresponding warehouse surplus, ensuring that the amount of feed is non-negative, and considering the upper limit of the food intake of sheep at a specific growth stage;

[0036] Step S65: By minimizing the objective function and satisfying all constraints at the same time, the feeding amount of each feed for the growth stage to be recommended is obtained.

[0037] As a preferred embodiment, step S7 includes:

[0038] Step S71: The system provides a user query interface, allowing the user to query the Hu sheep feeding plan according to actual needs; the system regularly collects actual data on Hu sheep feeding, including feed dosage, growth status, feed cost, etc., and compares it with the expected effect to form feedback data;

[0039] Step S72: Based on the feedback data, the system updates the knowledge graph of Hu sheep feeding, including adding new feed types, updating feeding plans, and revising nutritional demand predictions;

[0040] Step S73: Using the new feedback data, the system trains and optimizes the constrained optimization algorithm and case-based reasoning model to improve the accuracy and adaptability of feeding recommendations, and learns and iterates by continuously collecting new data; the system provides users with feeding recommendations based on user feedback and query history, including recommendations for specific growth stages, specific feed preferences, etc., to continuously improve the intelligence level of feeding recommendations.

[0041] As a preferred embodiment: feed formula data includes feed, feed type, feed price, daily nutritional requirements of Hu sheep, feed nutritional content, feed surplus in warehouse, and growth stage of Hu sheep.

[0042] Beneficial effects of the present invention:

[0043] A Hu-sheep precision feeding model based on knowledge graph and case-based reasoning is proposed. The model uses a case-based reasoning data prediction method to quantitatively supplement the Hu-sheep precision feeding recommendation method based on knowledge graph. By integrating data mining algorithms in the case-based reasoning stage, the method based solely on knowledge graphs has improved the situation that the data utility is not fully mined and the numerical prediction problem is not good at being processed. The proposed model can obtain highly accurate recommendation results by inputting less real-time data from sheep farms. The presented results also include qualitative feeding plans and nutritional requirements, with high integrity, and the recommendation process has strong interpretability, which has a driving effect on the rational use of feed in the growth process of Hu-sheep and the refined development of animal husbandry. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Attached Figure 1 :A data and knowledge-driven precision feeding method for Hu sheep DETAILED DESCRIPTION

[0045] The following will combine the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention, a method for accurate feeding of Hu sheep based on dual drive of data and knowledge, an embodiment of the method includes the following steps:

[0046] Step S1: The basic information of different types of feed is obtained from the National Agricultural Science Data Center, from which the types of feed suitable for feeding Hu sheep are screened, and then the basic information of various feeds is automatically extracted based on templates, and finally the obtained information is manually sorted and corrected to obtain the Hu sheep feed data set;

[0047] The feeding guidance opinions are based on the data source of "Huyang Breeding Technology" written by Huzhou Academy of Agricultural Sciences. The feeding guidance opinions are automatically extracted based on templates, and finally the obtained information is manually sorted and corrected to obtain the Huyang feeding guidance data set;

[0048] Only the feeding experiment record papers closely related to Hu sheep in CNKI were used as the data source of the cases. The papers on Hu sheep experiment record were searched in CNKI, and then information was extracted from the obtained papers to obtain the collection of Hu sheep feeding cases. The cases with clear values ​​of various indicators and complete recording processes were retained and stored in the case library, thus obtaining the Hu sheep feeding case data set.

[0049] The above data sets are integrated and stored in a relational database to form a basic knowledge graph of Hu sheep feeding.

[0050] Step S2: When there is a new feeding event, in the feeding plan recommendation part based on the knowledge graph, firstly, the knowledge graph representation learning is performed on the Hu sheep feeding knowledge graph, and then the feeding plan for the growth stage to be fed is predicted by the vector obtained by the representation learning, combined with the existing information in the graph as the result of knowledge graph reasoning. Among them, step S2 includes:

[0051] Step S21: performing information retrieval on the knowledge related to the growth stage of the Hu sheep to be fed, comparing the retrieved information types with the information types required for the recommendation results, and obtaining the missing information types in the knowledge graph;

[0052] In the embodiment of the present invention, information retrieval is performed on the growth stage-related knowledge of the Hu sheep to be fed, and the specific retrieval content covers growth stage characteristics, nutritional requirements, common feeding problems, etc. The types of information retrieved are compared in detail with the types of information required for the recommendation results, and the types of information missing in the knowledge graph are identified through automated scripts or manual review, such as feed types for specific growth stages, recommended feeding amounts, etc.

[0053] Step S22: using the knowledge graph representation learning model to map entities and relationships into low-dimensional vectors, using the numerical entity and relationship vectors for inference and prediction, and obtaining the value of the missing information; wherein step S22 includes:

[0054] Step S221: According to step S1 of claim 1, construct a knowledge graph of Hu sheep feeding, using a knowledge reasoning model that can fully capture the semantic connections between entities;

[0055] In the embodiment of the present invention, according to step S1 of claim 1, the collected data related to feeding of Hu sheep are used to construct a knowledge graph of feeding of Hu sheep, including entities such as "Hu sheep growth stage", "Hu sheep growth stage feeding management", "Hu sheep daily nutritional requirements", "feed", "feed nutrition", "feed price", "feed warehouse balance", and relationships such as "suitable for", "nutritional requirements", "belong to", etc. A knowledge reasoning model that can fully capture the semantic connection between entities, such as TransE, RotatE, PairRE, etc., is selected for subsequent representation learning.

[0056] Step S222: Based on the representation vectors of each entity and relationship in the graph obtained by representation learning, the growth stage to be fed is used as the head entity, and for each missing relationship, each entity in the graph is used as the tail entity to calculate the scoring function;

[0057] In the embodiment of the present invention, a representation learning model is used to map entities and relationships in the knowledge graph into low-dimensional vectors. The growth stage to be fed is taken as the head entity, and for each missing relationship, each entity in the graph is taken as the tail entity, and the scoring function between them is calculated. The higher the score, the stronger the association between the entity and the head entity.

[0058] Step S223: when the feeding plan for Hu sheep obtained through information retrieval is incomplete, the missing information is supplemented using link prediction technology;

[0059] In an embodiment of the present invention, when the feeding plan for Hu sheep obtained through information retrieval is incomplete, link prediction technology is applied to predict and complete the missing information based on existing entities and relationship vectors, such as inferring the appropriate concentrate-to-roughage ratio of feed at a certain growth stage and its supplementary feeding conditions.

[0060] Step S23: combining the existing information in the atlas, and taking the predicted feeding plan and the existing feeding plan in the atlas as the recommendation result.

[0061] Preferably, using the PairRE model, entity and relationship mapping refers to embedding the head entity into a vector h, the tail entity into a vector t, and the relationship into a vector r. H and r T Two vectors.

[0062] Use numerical entity and relationship vectors for inference and prediction to obtain the value of missing information. Combined with the existing information in the graph, the predicted feeding plan and the existing feeding plan in the graph are used as the recommendation result.

[0063] The inference prediction refers to the expectation that for a real triplet, ( is the Hadamard product), otherwise, the expectation keep away

[0064] Step S3: Divide the case as a whole into a problem description part and a solution part. According to step S1 of claim 1, the data source is the papers on Huyang experiment records in CNKI, and the values ​​of various indicators of Huyang feeding cases are obtained by using automated information extraction technology combined with manual sorting. Step S3 includes:

[0065] Step S31: The indicators of the problem description part include the growth stage of Hu sheep, feeding management, feed nutrition, metabolizable energy, dry matter, crude protein, neutral detergent fiber, acid detergent fiber, crude fat, crude fiber, crude ash, calcium and phosphorus;

[0066] Step S32: The solution part includes feeding environment requirements, feed type, feed content in the formula, and feed concentrate and coarse ratio;

[0067] Step S33: The values ​​of various indicators in the Hu sheep feeding case include metabolizable energy, dry matter, crude protein, neutral detergent fiber, acid detergent fiber, crude fat, crude fiber, crude ash, calcium and total phosphorus, which refer to ME, DM, CP, NDF, ADF, EE, CF, Ash, Ca and P respectively.

[0068] In the embodiment of the present invention, the indicators of the problem description part and the solution part are clarified, and the relevant indicator values ​​of the Hu sheep feeding case are extracted from data sources such as China National Knowledge Infrastructure (CNKI) by using automated information extraction technology combined with manual sorting to ensure the accuracy and completeness of the data.

[0069] Step S4: k groups of historical cases that are relatively close to the event to be recommended are selected as the case retrieval results based on the similarity between the event to be recommended and the problem description of each case in the case library. Step S4 includes:

[0070] Step S41: In the attribute similarity calculation, the attributes are divided into two types according to the data type: text attributes and numerical attributes. The text attributes include the growth stage of Hu sheep and feeding management, and the numerical attributes include feed nutrition, metabolizable energy, dry matter, crude protein, neutral detergent fiber, acid detergent fiber, crude fat, crude ash, calcium and total phosphorus;

[0071] Step S42: the similarity of the same numerical attribute between different cases can be represented by the distance between the two values ​​after normalization;

[0072] Step S43: using the distributed low-dimensional vectors of each growth stage obtained after the knowledge graph representation learning as the numerical representation of each growth stage, and calculating the similarity between the growth stages using the cosine similarity between the vectors;

[0073] Step S44: Use the formula to calculate the global similarity of the cases, and use the k historical cases with the highest global similarity and the lowest feed-to-meat ratio as the results of case retrieval.

[0074] Step S5: The k similar cases obtained in the case retrieval stage are used to modify the retrieved cases, and the similarity is applied to the prediction of the daily required nutrition. The predicted values ​​obtained after adjustment of each similar case are summed up, and then divided by the sum of the similarities to obtain the required nutrition for the growth stage of the event to be recommended.

[0075] The required nutritional requirement is obtained through the global similarity between these k cases and the event to be recommended.

[0076]

[0077] Where S(A l ,A o ) is the event to be recommended A o Similar to Case A l The global similarity between them, k is the number of similar cases retrieved, d l is the value of the solution for the similar case, and d is the predicted value.

[0078] Step S6: Using a feed feeding amount acquisition algorithm to obtain the most efficient feed formula for the existing data, wherein the data includes feed, feed type, feed price, daily nutritional requirements of Hu sheep, feed nutritional content, warehouse feed surplus, and Hu sheep growth stage.

[0079] Wherein, step S6 further comprises:

[0080] Step S61: define problem variables, including feed usage vector, feed price vector, Hu sheep nutritional demand vector, feed nutritional content matrix and warehouse feed surplus vector, etc.;

[0081] Step S62: Calculate the total cost, that is, the sum of the product of the amount of each feed and its corresponding price;

[0082] Step S63: determining an objective function, aiming to minimize the total cost;

[0083] Step S64: setting constraints, including satisfying the nutritional needs of the Hu sheep, ensuring that the amount of each feed is less than the corresponding warehouse surplus, ensuring that the amount of feed is non-negative, and considering the upper limit of the food intake of sheep at a specific growth stage;

[0084] Step S65: By minimizing the objective function and satisfying all constraints at the same time, the feeding amount of each feed for the growth stage to be recommended is obtained.

[0085] Preferably, the feed feeding amount acquisition algorithm is a cost minimization problem, that is, the goal is to minimize the total cost The four constraints include that the nutritional needs of Huyang sheep should be met. j=1,2,…,m; the amount of each feed must be less than the corresponding warehouse surplus x i <G i ,i=1,2,…,n; the amount of feed should be a non-negative number x i ≥0,i=1,2,…,n; There is an upper limit to the food intake of sheep at a certain growth stage r=1,2,…,k.

[0086] In the embodiment of the present invention, the constrained optimization algorithm can quickly and effectively find the optimal solution, avoid the cumbersome process of manual trial and error, and can flexibly respond to different constraints and objective functions, such as feed inventory, nutritional requirements, etc., to ensure that the formulated plan conforms to the actual situation. In formulating the feeding plan, the constrained optimization algorithm can handle these complex factors and constraints in a systematic way, and find the optimal feeding plan to minimize the feeding cost while ensuring that the lake sheep obtains sufficient nutritional support. The efficiency and global search capability of the algorithm ensure that the best solution can be quickly found under large-scale data and constraints, and its adjustability and interpretability also make the implementation and adjustment of the plan more flexible and controllable. Therefore, the use of the constrained optimization algorithm can effectively improve the efficiency and accuracy of the formulation of the feeding plan, thereby optimizing the economic benefits and nutritional support of lake sheep farming.

[0087] Step S7: Evaluate the feeding effect regularly and adjust and optimize the feeding plan according to the evaluation results.

[0088] In an embodiment of the present invention, the system provides a user query interface, allowing the user to query the feeding plan of Hu sheep according to actual needs; the system regularly collects actual data on Hu sheep feeding, including feed usage, growth conditions, feed costs, etc., and compares it with the expected effects to form feedback data; based on the feedback data, the system updates the Hu sheep feeding knowledge graph, including adding new feed types, updating feeding plans, and revising nutritional demand forecasts, etc.; using the new feedback data, the system trains and optimizes the constrained optimization algorithm and case reasoning model to improve the accuracy and adaptability of feeding recommendations, and learns and iterates by continuously collecting new data; the system provides users with feeding recommendations based on user feedback and query history, including recommendations for specific growth stages, specific feed preferences, etc., to continuously improve the intelligence level of feeding recommendations.

[0089] Through precision feeding methods, the amount of feed can be accurately measured and controlled to ensure that the Hu sheep obtain the appropriate amount of nutrition while avoiding waste and overfeeding. Precision feeding methods can also improve the feeding efficiency of Hu sheep, reduce the feed-to-meat ratio, and reduce the feed cost of Hu sheep feeding.

[0090] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A data and knowledge-driven precision feeding method for Hu sheep, characterized in that: The following steps are involved: Step S1: After data retrieval and extraction processing, a Hu sheep feed data set, a Hu sheep feeding guidance data set, and a Hu sheep feeding case data set are obtained; each data set is integrated and stored in a relational database to form a basic Hu sheep feeding knowledge graph; Step S2: When there is a new feeding event, the knowledge graph representation learning is performed on the Hu sheep feeding knowledge graph, and then the feeding plan for the growth stage to be fed is predicted by the vector obtained by representation learning, combined with the existing information in the graph as the result of knowledge graph reasoning; Step S3: Divide the overall case of Huyang feeding into a problem description part and a solution part; Step S4: Filter out k groups of historical cases that are relatively close to the event to be recommended as the result of case retrieval based on the similarity between the event to be recommended and the problem description of each case in the case library; Step S5: through the k similar cases obtained in the Huyang feeding case retrieval stage, the retrieved cases are corrected, and the global similarity between the event to be recommended and the similar cases is applied to the prediction of daily nutrition requirements; The predicted values ​​obtained after adjustment for each similar case are summed up, and then divided by the sum of similarities to obtain the required nutrients for the growth stage of the event to be recommended; Step S6: using a feed feeding amount acquisition algorithm to obtain the most efficient feed formula for the existing data; Step S7: Evaluate the feeding effect regularly and adjust and optimize the feeding plan according to the evaluation results.

2. The method according to claim 1, characterized in that: The Hu sheep feeding case collection and data set are based on the feeding experiment record papers closely related to Hu sheep in CNKI as the data source of the cases. Hu sheep experiment record papers are searched in CNKI, and then information is extracted from the obtained papers. Cases with clear numerical values ​​of various indicators and complete recording processes are retained and stored in the case library to obtain the Hu sheep feeding case data set.

3. The method according to claim 1 or 2, characterized in that: The Hu sheep feed data set, Hu sheep feeding instruction data set, and Hu sheep feeding case data set are all obtained through template-based automatic information extraction and manual sorting and correction.

4. The method according to claim 1, characterized in that in, Step S2 includes: Step S21: performing information retrieval on the knowledge related to the growth stage of the Hu sheep to be fed, comparing the retrieved information types with the information types required for the recommendation results, and obtaining the missing information types in the knowledge graph; Step S22: using the knowledge graph representation learning model to map entities and relationships into low-dimensional vectors, using the numerical entity and relationship vectors for inference and prediction, and obtaining the value of the missing information; wherein step S22 includes: Step S221: According to step S1, construct a knowledge graph of Hu sheep feeding, using a knowledge reasoning model that can fully capture the semantic connections between entities; Step S222: Based on the representation vectors of each entity and relationship in the graph obtained by representation learning, the growth stage to be fed is used as the head entity, and for each missing relationship, each entity in the graph is used as the tail entity to calculate the scoring function; Step S223: when the feeding plan for Hu sheep obtained through information retrieval is incomplete, the missing information is supplemented using link prediction technology; Step S23: combining the existing information in the atlas, and taking the predicted feeding plan and the existing feeding plan in the atlas as the recommendation result.

5. The method according to claim 1, characterized in that in, Step S3 includes: Step S31: The indicators of the problem description part include the growth stage of Hu sheep, feeding management, feed nutrition, daily nutrient requirements, metabolizable energy, dry matter, crude protein, neutral detergent fiber, acid detergent fiber, crude fat, crude fiber, crude ash, calcium and phosphorus; Step S32: The solution part includes feeding environment requirements, feed type, feed content in the formula, and feed concentrate and coarse ratio; Step S33: The values ​​of various indicators in the Hu sheep feeding case include metabolizable energy, dry matter, crude protein, neutral detergent fiber, acid detergent fiber, crude fat, crude fiber, crude ash, calcium and total phosphorus, which refer to ME, DM, CP, NDF, ADF, EE, CF, Ash, Ca and P respectively.

6. The method according to claim 1, characterized in that in, Step S4 includes: Step S41: In the attribute similarity calculation, the attributes are divided into two categories according to the data type: text attributes and numerical attributes; the text attributes include the growth stage of Hu sheep and feeding management, and the numerical attributes include feed nutrition, daily nutritional requirements, metabolizable energy, dry matter, crude protein, neutral detergent fiber, acid detergent fiber, crude fat, crude ash, calcium and total phosphorus; Step S42: the similarity of the same numerical attribute between different cases is represented by the distance between the two values ​​after normalization; Step S43: using the distributed low-dimensional vectors of each growth stage obtained after the knowledge graph representation learning as the numerical representation of each growth stage, and calculating the similarity between the growth stages using the cosine similarity between the vectors; Step S44: Use the formula to calculate the global similarity of the cases, and use the k historical cases with the highest global similarity and the lowest feed-to-meat ratio as the results of case retrieval.

7. The method according to claim 1, characterized in that in, Step S6 includes: Step S61: define problem variables, including feed usage vector, feed price vector, Hu sheep nutritional demand vector, feed nutritional content matrix and warehouse feed surplus vector, etc.; Step S62: Calculate the total cost, that is, the sum of the product of the amount of each feed and its corresponding price; Step S63: determining an objective function, aiming to minimize the total cost; Step S64: setting column constraints, including satisfying the nutritional needs of Hu sheep, ensuring that the amount of each feed is less than the corresponding warehouse surplus, ensuring that the amount of feed is non-negative, and considering the upper limit of the food intake of sheep at a specific growth stage; Step S65: By minimizing the objective function and satisfying all constraints at the same time, the feeding amount of each feed for the growth stage to be recommended is obtained.

8. The method according to claim 1, characterized in that in, Step S7 includes: Step S71: The system provides a user query interface, allowing the user to query the Hu sheep feeding plan according to actual needs; the system regularly collects actual data on Hu sheep feeding, including feed dosage, growth status, feed cost, etc., and compares it with the expected effect to form feedback data; Step S72: Based on the feedback data, the system updates the knowledge graph of Hu sheep feeding, including adding new feed types, updating feeding plans, and revising nutritional demand predictions; Step S73: Using the new feedback data, the system trains and optimizes the constrained optimization algorithm and case-based reasoning model to improve the accuracy and adaptability of feeding recommendations, and learns and iterates by continuously collecting new data; the system provides users with feeding recommendations based on user feedback and query history, including recommendations for specific growth stages, specific feed preferences, etc., to continuously improve the intelligence level of feeding recommendations.

9. The method according to claim 1, characterized in that: The feed formula data include feed, feed type, feed price, daily nutritional requirements of Hu sheep, feed nutritional content, warehouse feed surplus, and Hu sheep growth stage.

Images