Construction of Rice Fertilization Knowledge Graph Considering Spatiotemporal Features and Knowledge Inference Method

Abstract

The present invention relates to a method for constructing a knowledge graph of rice fertilization and knowledge reasoning that takes into account spatio-temporal characteristics. The method: First, knowledge modeling is carried out using a six-tuple structure to construct a rice fertilization ontology, which includes four core elements: rice, planting environment, nutrient element diagnosis, and fertilization plan. Among them, the spatio-temporal differences of rice fertilization knowledge are expressed jointly through spatio-temporal relationships, spatial attributes, and time states; Secondly, on this basis, a method for reasoning rice fertilization strategies based on a knowledge graph is proposed. Combining the certainty factor model, nutrient element diagnosis is realized through knowledge reasoning, and fertilization plans for different stages are formed based on historical case retrieval and the nutrient balance method. Finally, knowledge extraction is carried out on the spatio-temporal data of crowd-sourced perception related to rice fertilization to form a knowledge graph of rice fertilization for practice. The method of the present invention can realize the expression and reasoning of fertilization decisions that take into account the rice growth process and complex geographical environment, and provide support for fertilizing rice according to local conditions and seasons.

Description

Technical Field

[0001] The present invention belongs to the field of information technology, and particularly relates to a method for constructing a rice fertilization knowledge graph considering spatio-temporal characteristics and knowledge reasoning. Background Art

[0002] As one of the world's three major food crops, the yield and quality of rice are crucial for global food security. Fertilization helps to balance soil nutrients, improve crop yield and quality [1], but blind fertilization will lead to a series of problems such as yield decline, low fertilizer utilization rate, and environmental pollution [2]. In order to reduce the negative impact on the environment while ensuring the rice yield, it is necessary to develop precision agriculture to guide farmers to use fertilizers scientifically. However, due to the complex mechanism and significant spatio-temporal heterogeneity of rice fertilization, the cost of expert fine-grained planting guidance is high, and farmers lack expert knowledge to guide fine-grained planting and cannot use fertilizers scientifically according to different spatio-temporal conditions. Therefore, it is necessary to consider factors such as time and space, and integrate expert knowledge to make fertilization decisions under complex spatio-temporal conditions, so as to provide support for promoting scientific fertilizer use in rice and realizing sustainable agricultural development.

[0003] With the promotion of the development of big data technology, the amount of agricultural data has increased explosively, showing the characteristics of huge data volume, wide sources, and complex structure, which contains rich agricultural knowledge and spatio-temporal characteristics. This puts forward higher requirements for the integration, processing, and analysis capabilities of agricultural data and information resources. At present, the research on rice fertilization decision-making mainly uses mathematical models and expert models to calculate the fertilization amount. Expert models [3, 4] generally use the method of setting based on agronomic mechanisms and empirical rules, which has a certain interpretability, but has weak utilization ability for big data, cannot be dynamically updated, and is difficult to handle complex reasoning [5]. Mathematical models are generally based on methods such as statistical regression [6], machine learning [7], and deep learning [8], which are objective but have poor interpretability and cannot meet the needs of low-cost crop nutrient element diagnosis and precise plan recommendation based on spatio-temporal characteristics.

[0004] Knowledge graph is a knowledge system that structurally describes concepts, entities, and the relationships between them [9, 10]. Some studies have shown that knowledge graphs have great advantages in integrating and organizing multi-source heterogeneous data [11, 12], which helps to form an intelligent knowledge service for efficient association and integrated utilization

[13] . In the agricultural field, effectively organizing agricultural knowledge is the basis for realizing agricultural knowledge services such as intelligent diagnosis and solution recommendation. Some researchers have achieved the effective organization of agricultural knowledge by constructing agricultural ontologies and knowledge graphs. For example, the Food and Agriculture Organization of the United Nations (FAO) led the construction of the agricultural ontology AGROVOC

[14] , realizing knowledge modeling and sharing of more than 40,000 concepts and more than 9.6 million terms in the fields of agriculture, fisheries, forestry, and the environment. Zhou Jun et al.

[15] proposed a rice fertilization knowledge extraction model and constructed a rice fertilization knowledge graph to achieve the acquisition and expression of agricultural knowledge for multi-source heterogeneous agricultural data. The School of Data Science and Engineering of East China Normal University led the construction of the agricultural knowledge graph AgriKG

[16] , and on this basis, developed an intelligent question-answering system for smart agriculture to achieve efficient retrieval of agricultural knowledge.

[0005] At present, agricultural knowledge graphs have achieved good application and development in semantic search, intelligent question answering, intelligent diagnosis, and solution recommendation [17, 18]. Since agricultural decision-making applications such as intelligent diagnosis and solution recommendation are of great significance to actual agricultural production, some scholars in the agricultural field have focused on this and carried out research on knowledge reasoning techniques and methods for realizing intelligent diagnosis and solution recommendation based on knowledge graphs. In terms of agricultural intelligent diagnosis, the research mostly adopts the method of combining knowledge graphs with mathematical models or expert models. Wang et al.

[19] constructed a dairy disease knowledge graph, used a BiLSTM-CNN hybrid network to learn the semantic features between diseases and symptoms in the knowledge graph, and realized the diagnosis of dairy diseases according to symptoms. Yu Helong et al.

[20] constructed a rice pest and disease knowledge graph and realized the diagnosis of rice pests and diseases based on symptom entities combined with the certainty factor model. In terms of agricultural solution recommendation, Gu et al.

[21] constructed a tobacco disease prevention and control model based on case-based reasoning and knowledge graphs, and matched tobacco disease prevention and control solutions through environmental indicators. Ge Weixi et al.

[22] proposed a rice precise fertilization recommendation model based on knowledge graphs and case-based reasoning, considering rice varieties and soil environments to recommend the overall rice fertilization plan, but unable to recommend fertilization plans for different growth periods.

[0006] The main disadvantages of the existing technologies are:

[0007] (1) The existing research on rice fertilization knowledge graphs lacks the modeling of the spatio-temporal characteristics of knowledge, and there is a problem of insufficient expression of the spatio-temporal differences in rice fertilization knowledge.

[0008] (2) In existing research, rice nutrition diagnosis models make decisions based only on a single symptom element, unable to obtain the spatial characteristics of the rice planting environment, lacking full utilization of rice fertilization knowledge, and having the problem of not considering the complex relationship between spatio-temporal characteristics and rice nutritional disorders.

[0009] (3) In existing research, rice fertilization plan recommendation models mainly use environmental factors to recommend fertilization amounts, but lack consideration of the impact of rice growth stages and actual planting conditions on the fertilization plan, and are unable to comprehensively utilize spatio-temporal characteristics and historical cases to recommend refined fertilization plans.

[0010] References:

[0011] [1]. St. Clair, S. B. and J. P. Lynch, The opening of Pandora’s Box: climate change impacts on soil fertility and crop nutrition in developing countries. Plant and Soil, 2010. 335(1): p. 101 - 115.

[0012] [2]. Xu Yang et al., Progress and prospects of the soil testing and formulated fertilization project in the past fifteen years. China Soil and Fertilizer, 2023. No. 311(03): pp. 236 - 244.

[0013] [3]. Wang Yilun et al., Research on the effects of recommended fertilization of wheat - maize based on the nutrient expert system. Scientia Agricultura Sinica, 2015. 48(22): pp. 4483 - 4492.

[0014] [4]. H., K., K. S. S. and D. C. R. Expert system for diagnosis of diseases of rice plants: Prototype design and implementation. in 2016 International Conference on Automatic Control and Dynamic Optimization Techniques (ICACDOT). 2016.

[0015] [5]. Liu X, Bai X, Wang L, et al. Review and Trend Analysis of Knowledge Graphs for Crop Pest and Diseases[J]. IEEE Access. 2019, 7: p. 62251 - 62264.

[0016] [6]. WANG, S.N., J.K. CHENG and Y.C. LIAO, FERTILIZATION MODEL FOR FLUE - CURED TOBACCO (NICOTIANA TABACUM L.) IN SOUTHWEST CHINA. Applied Ecology and Environmental Research, 2020. 18(6): p. 7853 - 7863.

[0017] [7]. Gao, J., et al., A Fertilization Decision Model for Maize, Rice, and Soybean Based on Machine Learning and Swarm Intelligent Search Algorithms. Agronomy - Basel, 2023. 13(5): p. 1400.

[0018] [8]. Escalante, H.J., et al., Barley yield and fertilization analysis from UAV imagery: a deep learning approach. International Journal of Remote Sensing, 2019. 40(7): p. 2493 - 2516.

[0019] [9]. Nickel, M., et al., A Review of Relational Machine Learning for Knowledge Graphs. Proceedings of the IEEE, 2015. 104(1): p. 11 - 33.

[0020]

[10] . Ballatore, A., M. Bertolotto and D. C. Wilson, A Structural-Lexical Measure of Semantic Similarity for Geo-Knowledge Graphs. ISPRS INTERNATIONAL JOURNAL OF GEO-INFORMATION, 2015. 4(2): p. 471-492.

[0021]

[11] . Yan, J., et al., A retrospective of knowledge graphs. Frontiers of Computer Science, 2018. 12(1): p. 55-74.

[0022]

[12] . Lu Feng, Yu Li and Qiu Peiyuan, On the geographical knowledge graph. Journal of Geo-Information Science, 2017. 19(06): pp. 723-734.

[0023]

[13] . Yang Deqing et al., Knowledge-driven recommendation systems: current status and prospects. Journal of Information Security, 2021. 6(05): pp. 35-51.

[0024]

[14] . Rajbhandari, S. and J. Keizer, The AGROVOC Concept Scheme: A Walkthrough. Journal of Integrative Agriculture, 2012. 11: p. 694-699.

[0025]

[15] . Zhou Jun et al., Research on information extraction of rice fertilization knowledge graph based on improved CASREL. Transactions of the Chinese Society for Agricultural Machinery, 2022. 53(11): pp. 314-322.

[0026]

[16] . Chen, Y., et al., AgriKG: An Agricultural Knowledge Graph and Its Applications, in Database Systems for Advanced Applications. DASFAA 2019. 2019, Springer International Publishing: Chiang Mai, Thailand. p. 533-537.

[0027]

[17] .Chicaiza,J.and P.Valdiviezo-Diaz,A Comprehensive Survey of Knowledge Graph-Based Recommender Systems: Technologies, Development, and Contributions, in Information. 2021.

[0028]

[18] .Johansson,J.,et al.,Supporting connectivism in knowledge based engineering with graph theory, filtering techniques and model quality assurance. Advanced Engineering Informatics, 2018. 38: p. 252-263.

[0029]

[19] .Wang H,Shen W,Zhang Y,et al.Diagnosis of Dairy Cow Diseases by Knowledge-Driven Deep Learning Based on the Text Reports of Illness State[J].Comput. Electron. Agric. 2023, 205: p. 107564.

[0030]

[20] . Yu Helong, et al., Intelligent Diagnosis System for Rice Diseases and Insect Pests Based on Knowledge Graph. Journal of South China Agricultural University, 2021. 42(05): pp. 105-116.

[0031]

[21] .Gu,L.,et al.,Research on the model for tobacco disease prevention and control based on case-based reasoning and knowledge graph. Filomat, 2018. 32: p. 1947-1952.

[0032]

[22] . Ge Weixi, et al., Precision Fertilization Recommendation Model for Rice Based on Knowledge Graph and Case-Based Reasoning. Transactions of the Chinese Society of Agricultural Engineering, 2023. 39(2): pp. 126-133. Summary of the Invention

[0033] The object of the present invention is to standardize the description of the fertilization mechanism and spatio-temporal relationship of rice, construct a rice fertilization knowledge graph considering spatio-temporal characteristics, and on this basis, propose a method for constructing a rice fertilization knowledge graph and knowledge reasoning considering spatio-temporal characteristics. This method combines the certainty factor model with the knowledge graph, realizes the diagnosis of rice nutrient elements by integrating the soil environment and appearance characteristics, introduces the historical case retrieval technology, recommends differential fertilization schemes by integrating the soil environment and rice growth stages, realizes the refined fertilization decision-making application of rice under different spatio-temporal conditions, and provides support for the fertilization of rice according to local conditions in different times.

[0034] To achieve the above object, the technical solution of the present invention is: a method for constructing a rice fertilization knowledge graph and knowledge reasoning considering spatio-temporal characteristics, including:

[0035] S1. Construct a rice fertilization knowledge graph considering spatio-temporal characteristics: design the ontology of the rice fertilization knowledge graph, complete the knowledge modeling of the rice fertilization knowledge graph considering spatio-temporal characteristics, and on this basis, perform rice fertilization knowledge extraction and knowledge storage;

[0036] S2. Propose a method for reasoning about rice fertilization strategies based on the knowledge graph, including the construction of a rice nutrient element diagnosis model and a rice fertilization scheme recommendation model;

[0037] S3. Use the spatio-temporal data of rice fertilization crowd-sensing as experimental data, and analyze and evaluate the reasoning results of rice fertilization strategies based on the constructed knowledge graph.

[0038] In an embodiment of the present invention, step S1 is specifically implemented as follows:

[0039] (1) Knowledge modeling of the rice fertilization knowledge graph considering spatio-temporal characteristics: using a top-down method, through ontology logical structure expression, concept system establishment, instance-state-attribute representation, and relationship determination, complete the knowledge modeling of the rice fertilization knowledge graph;

[0040] (2) Rice fertilization knowledge extraction and knowledge storage: for structured data, form the mapping from data tables to RDF according to the rice fertilization pattern ontology, and directly convert it into the expression of the knowledge graph; for semi-structured and unstructured data, first perform preprocessing operations on the data to obtain valid data, then perform template-based automated information extraction, and finally organize and correct the obtained information; obtain the time relationship through the discrimination and sorting of the extracted time elements, and use the nine-intersection model DE-9IM based on dimension expansion to describe the spatial relationship between geographical entity objects, as shown in the following formula:

[0041]

[0042] where, R DE-9IM(a,b)Represents the spatial topological relationship between geographical entities described by DE-9IM; a and b represent different rice geographical entities; I, B, and E represent the interior, boundary, and exterior of the geographical entity respectively; the dim function returns the maximum dimension of the intersecting geometric body;

[0043] Using the Neo4j graph database, based on the Resource Description Framework, various related elements of rice fertilization knowledge are represented in the form of triples (entity, relationship, entity) and (entity, attribute, attribute value). Subsequently, the Neo4j graph database is used to store rice fertilization knowledge in the "node-relationship" storage model. The graph database stores the entities and attribute values in the triples as nodes and the attributes and relationships as edges.

[0044] In an embodiment of the present invention, the specific implementation method of knowledge modeling of a rice fertilization knowledge graph considering spatio-temporal characteristics is as follows:

[0045] (a) Ontological logical structure expression of the rice fertilization knowledge graph

[0046] The ontology in the field of rice fertilization is abstracted into a six-tuple structure of concept, relationship, attribute, rule, instance, and state, and is formally represented as shown in the following formula:

[0047] Onto = (Con, Rel, Prop, Rule, Ins, Sta)

[0048] Among them, Con refers to the concept, which represents the general term of the set of things with the same characteristics in the rice fertilization knowledge system; Rel refers to the relationship, which represents the relationship types between rice fertilization concepts, between concepts and instances, and between instances; Prop refers to the attribute, which represents the semantic and spatio-temporal characteristics of different instance objects; Rule refers to the rule, which defines the constraint expressions of the value range, type, and combination method of concepts and instances in the field of rice fertilization to support semantic reasoning; Ins refers to the instance, which represents the specific expression based on the domain concept; Sta refers to the state, which represents the attribute characteristics of rice fertilization instances under different spatio-temporal conditions;

[0049] (b) Concept system establishment

[0050] Based on the ontology model of the rice fertilization knowledge graph, the rice fertilization concepts are sorted out, and a concept system for the field of rice fertilization including four core elements of rice, planting environment, nutrient element diagnosis, and fertilization plan is established;

[0051] (c) Instance-state-attribute representation

[0052] Guided by rice fertilization knowledge, define the instances, states, and attribute characteristics of the rice fertilization concept and its sub - concepts; the rice fertilization instance is a concrete expression of the concept, and the description of the rice attribute characteristics is centered around the instance. At the same time, to represent the characteristics of fertilization knowledge in terms of time or space, state nodes are set to divide the periodic characteristics of the instance;

[0053] (d) Relationship determination

[0054] Define the relationships between rice fertilization domains as time relationships, space relationships, semantic relationships, and mathematical relationships; the time relationship of rice fertilization refers to the sequence of occurrence between rice fertilization instances, expressed by time topological relationships; the space relationship of rice fertilization aims to describe the space relationship between geographical entity objects in the planting environment, and its space relationship is a space topological relationship, indicating the proximity and association degree between geographical entity objects; the semantic relationships of rice fertilization include inheritance relationships, instance relationships, and attribute relationships. The inheritance relationship represents the hierarchical structure between the upper and lower concepts of rice fertilization; the instance relationship represents the relationship between the concept and the instance; the attribute relationship is the expression of the mutual connection between entities at the semantic level and also the specific description of the relationship between the entity and the attribute value; the mathematical relationship of rice fertilization represents the correlation relationship formed by calculating between concepts or attributes through a mathematical model.

[0055] In an embodiment of the present invention, step S2 is specifically implemented as follows:

[0056] (1) Construct a rice nutrient element diagnosis model: Combine the certainty factor CF model with the knowledge graph, make full use of the rice appearance characteristics and planting plots in different planting areas, match the damage symptoms and soil environment elements with spatial characteristics in the rice fertilization spatio - temporal knowledge graph, obtain the CF values that can support multiple rule diagnosis conclusions, and calculate the types and probabilities of possible nutrient disorders that rice may suffer from;

[0057] (2) Construct a rice fertilization plan recommendation model: Based on the rice fertilization knowledge graph, integrate the fertilization experience knowledge of multiple rice varieties and the principle of nutrient - balanced fertilization, combine the similar historical fertilization case retrieval method, and make full use of rice variety, planting plot, and growth period information to recommend a refined fertilization plan including fertilization amount, fertilizer type, and fertilization time for rice under different environmental characteristics and growth periods.

[0058] In an embodiment of the present invention, the specific implementation method of constructing a rice nutrient element diagnosis model is as follows:

[0059] First, taking the appearance characteristics of rice and the planting plots in different planting areas as inputs, match the damage symptoms and soil environmental factors with spatial characteristics in the rice fertilization knowledge graph; secondly, further match the certainty factors of the damage symptoms and the soil environment based on the expert knowledge in the knowledge graph, and use the following two formulas to calculate the incidence probability of each nutritional disorder in the current rice planting situation, and take the nutritional disorder type corresponding to the highest incidence probability as the diagnosis result;

[0060] CF(B,A) = CF(B,A1) + CF(B,A2) - CF(B,A1) × CF(B,A2)

[0061]

[0062] Among them, A and B respectively represent evidence and conclusion, A1 and A2 are sub-evidences that make up evidence A, CF(B,A) represents the certainty factor, also known as the CF value; if CF(B,A) > 0, the larger the value of CF(B,A), the more the appearance of the evidence supports the conclusion to be true; if CF(B,A) < 0, the smaller the value of CF(B,A), the more the appearance of the evidence supports the conclusion to be false; if CF(B,A) = 0, the appearance of the evidence has no supporting effect on the conclusion; P(B|A) represents the diagnostic probability, and P(B) represents the prior probability of the conclusion.

[0063] In an embodiment of the present invention, the specific implementation method for constructing a rice fertilization plan recommendation model is as follows:

[0064] First, input information on rice variety, planting plot, and growth period, match the problem description index information through the rice fertilization knowledge graph, and obtain the spatio-temporal characteristics of the input information; then use the method of knowledge representation learning to obtain the low-dimensional vector representations of all entities and relationships in the rice fertilization knowledge graph, and calculate the global similarity and retrieve similar historical fertilization cases on this basis; when the similarity of the retrieval result is relatively high, generate a rice fertilization plan; when there are no similar historical fertilization cases, combine the existing information in the rice fertilization knowledge graph and use the calculation method of the nutrient balance method to supplement the specific fertilization amount value; finally, match the rice growth period with the fertilization plan to form the final rice fertilization plan;

[0065] (a) Retrieval of similar historical fertilization cases

[0066] The goal of retrieving similar historical fertilization cases is to match historical cases with relatively high similarity in the problem description part of the case library according to the rice planting situation; according to the data type, the attributes of the problem description part are divided into two categories: numerical attributes and text attributes;

[0067] Numerical attributes include organic matter, available nitrogen, available phosphorus, available potassium, pH value, and yield level; the similarity of the same numerical attribute among different cases is represented by the distance after normalizing two values, and the specific calculation process is shown in the following formula:

[0068]

[0069] In the formula and respectively represent the values of attribute k of the i-th case and the case j to be recommended, and respectively represent the maximum and minimum values of attribute k, represents and the numerical attribute similarity;

[0070] The text attributes are rice varieties and growth periods; when conducting similar historical fertilization cases, it is necessary to calculate the similarity between different rice varieties. Since the similarity of text attributes cannot be directly measured by a computer, it is necessary to represent them in a vectorized form; text attributes are stored in the form of entities and relationships in the knowledge graph, using discrete and explicit symbolic representations. Vector operations based on the vectorized representation of the knowledge graph can reflect the similarity between entities; the attribute relationships in the rice fertilization knowledge graph are all one-to-one relationships. Therefore, the TransE model is used for knowledge representation learning. The core idea of the TransE model is that when the triple (h, r, t) holds, it should satisfy the following formula:

[0071] h + r ≈ t

[0072] The definition of the model scoring function is shown in the following formula:

[0073] f r (h, t) = ||h + t - r|| L1 / L2

[0074] where h is the head entity, r is the relationship, t is the tail entity, and L1 and L2 are the first and second normal forms respectively; for a correct triple, it is expected that the score is smaller; for an incorrect triple, it is expected that the score is higher;

[0075] Taking the distributed low-dimensional vector of rice varieties obtained after knowledge graph representation learning as the numerical representation of various types of rice varieties, the similarity between rice varieties is calculated using the cosine similarity between vectors. The cosine similarity calculation formula is as follows:

[0076]

[0077] where and Vectors representing the rice variety attribute k of the case to be recommended and the i-th case respectively Indicates the cosine similarity between rice variety vectors;

[0078] The global similarity of cases is calculated using the following formula, and the historical cases with higher global similarity are used as the results of case retrieval:

[0079]

[0080] Among them, K represents the number of attributes, w k Represents the weight of attribute k, s k Represents the similarity of attribute k between two cases, S i Represents the global similarity between the i-th case and the case to be recommended;

[0081] (b) Calculation by nutrient balance method

[0082] The nutrient balance method aims to achieve the supply-demand balance of nutrients between crops and soil, and calculates the amount of fertilizer required to achieve the target yield according to the difference between the amount of fertilizer required by the crop and the amount of fertilizer that the soil can provide. The calculation process of the recommended fertilization amount is shown in the following formula:

[0083]

[0084] Among them, Fer refers to the required fertilization amount, Y represents the target yield, U represents the fertilizer utilization rate in the current season, W is the nutrient absorption amount per unit yield, C represents the nutrient content in the fertilizer, X represents the soil test value, and K represents the correction coefficient of soil available nutrients.

[0085] In an embodiment of the present invention, step S3 is specifically implemented as follows:

[0086] (1) Perform model performance testing on 5 common rice nutrient disorders, and evaluate the rice nutrient element diagnosis model using the accuracy evaluation index;

[0087] (2) Compare the recommended results of the current rice fertilization plan in the experimental area with the experimental results of the rice fertilization plan recommendation model, including the comparison of three types of results: fertilization period, fertilization amount, and fertilizer type.

[0088] Compared with the prior art, the present invention has the following beneficial effects:

[0089] (1) The present invention sorts out the knowledge of rice fertilization experts, integrates multi-source heterogeneous agricultural spatio-temporal data, effectively organizes and integrates complex rice fertilization knowledge in time and space dimensions, reduces the cost of obtaining agricultural expert knowledge, and provides support for rice fertilization decision-making under different spatio-temporal conditions.

[0090] (2) Fertilization decisions are made through knowledge reasoning based on the rice fertilization knowledge graph, which can provide differentiated solutions by comprehensively considering the spatial characteristics of the planting environment, the temporal characteristics of rice growth, and the growth conditions. While improving the refinement of fertilization plan recommendations, the interpretability of the model method is also taken into account through the visual display of the reasoning results. Description of the Drawings

[0091] Figure 1 It is the rice fertilization concept system based on ontology.

[0092] Figure 2 It is an example of the node and relationship representation of instance-attribute-status.

[0093] Figure 3 It is the relationship classification in the field of rice fertilization.

[0094] Figure 4 It is the rice nutrient element diagnosis process based on knowledge reasoning.

[0095] Figure 5 It is the fertilization plan recommendation process based on the knowledge graph.

[0096] Figure 6 It is the visualization of fertilization plan "001".

[0097] Figure 7 It is the visualization of the rice nutrient element diagnosis knowledge graph.

[0098] Figure 8 It is the visualization of the rice plan recommendation knowledge graph.

[0099] Figure 9 It is the applicability score of the fertilization plan. Detailed Implementation Manner

[0100] Next, in combination with the drawings, the technical solutions of the present invention will be specifically described.

[0101] The present invention proposes a new method for constructing a rice fertilization knowledge graph and knowledge reasoning that takes into account spatio-temporal characteristics. This method uses a six-tuple structure (concept, relationship, attribute, rule, instance, status) for knowledge modeling. The constructed rice fertilization ontology includes four core elements: rice, planting environment, nutrient element diagnosis, and fertilization plan. At the same time, the spatio-temporal differences of rice fertilization knowledge are expressed jointly through spatio-temporal relationships, spatial attributes, and temporal states. On this basis, a rice fertilization strategy reasoning method based on the knowledge graph is proposed. The diagnosis of rice nutrient elements is realized through knowledge reasoning in combination with the certainty factor model, and fertilization plans at different stages are formed based on historical case retrieval and nutrient balance method. Finally, using the spatio-temporal data of rice fertilization crowd-sensing in a certain county, an application example of the rice fertilization knowledge graph is formed and evaluated. Its main contents include:

[0102] (1) Construct a rice fertilization knowledge graph that takes into account spatio-temporal features. Design the ontology of the rice fertilization knowledge graph, complete the knowledge modeling of rice fertilization that takes into account spatio-temporal features, and on this basis, perform rice fertilization knowledge extraction and knowledge storage;

[0103] (2) Propose a rice fertilization strategy reasoning method based on the knowledge graph, including two parts: rice nutrient element diagnosis and fertilization plan recommendation;

[0104] (3) Experimental analysis and evaluation. Use the spatio-temporal data of the rice fertilization crowd-sensing in a certain county as experimental data, and analyze and evaluate the results of the rice fertilization strategy reasoning based on the constructed knowledge graph.

[0105] The purpose of the present invention is to propose a method for constructing a rice fertilization knowledge graph and knowledge reasoning that takes into account spatio-temporal features. The specific steps are introduced as follows:

[0106] The first step is to construct a rice fertilization knowledge graph that takes into account spatio-temporal features

[0107] Knowledge modeling of rice fertilization that takes into account spatio-temporal features. Using a top-down approach, through ontology logical structure expression, concept system establishment, instance-state-attribute representation, and relationship determination, complete the knowledge modeling of the rice fertilization knowledge graph.

[0108] Rice fertilization knowledge extraction and knowledge storage. For structured data such as relational databases and statistical tables, according to the ontology of the rice fertilization pattern, form the mapping of the data table to RDF, and directly convert it into the expression of the knowledge graph. For semi-structured and unstructured data such as network texts and scientific and technological literature, which contain a large amount of redundant information, first preprocess the data to obtain effective data, then perform template-based automated information extraction, and finally manually organize and correct the obtained information. The time relationship is obtained by judging and sorting the extracted time elements, and the spatial relationship between geographical entity objects is described by using the Dimensionally Extended 9-Intersection Model (DE-9IM) based on dimension expansion, as shown in the following formula.

[0109]

[0110] Among them, RDE-9IM(a,b) represents the spatial topological relationship between geographical entities described by the DE-9IM model; a and b represent different rice geographical entities; I, B, and E represent the interior, boundary, and exterior of the geographical entity respectively; the dim function returns the maximum dimension of the intersecting geometric bodies.

[0111] Using the Neo4j graph database and based on the Resource Description Framework, various relevant elements of rice fertilization knowledge are represented in the form of triples of (entity, relationship, entity) and (entity, attribute, attribute value). Subsequently, the Neo4j graph database is used to store rice fertilization knowledge in a "node-relationship" storage model. The graph database stores the entities and attribute values in the triples as nodes and the attributes and relationships as edges.

[0112] Step 2: Construction of the rice fertilization strategy reasoning method based on the knowledge graph

[0113] The rice fertilization strategy reasoning method based on the knowledge graph includes two parts: rice nutrient element diagnosis and fertilization plan recommendation. The process of the rice nutrient element diagnosis model is as shown in the appendix Figure 4 shown, and the process of the fertilization plan recommendation model is as shown in the appendix Figure 5 shown.

[0114] (2) Rice nutrient element diagnosis model. Combining the certainty factor (CF) model with the knowledge graph, making full use of the appearance characteristics of rice and the planting plots in different planting areas, matching the damage symptoms and soil environment elements with spatial characteristics in the rice fertilization spatio-temporal knowledge graph, obtaining the CF values that can support the diagnostic conclusions of multiple rules, and calculating the types and probabilities of nutrient disorders that rice may suffer from;

[0115] (3) Rice fertilization plan recommendation model. Based on the rice fertilization knowledge graph, integrating the fertilization experience knowledge of multiple rice varieties and the principle of balanced nutrient fertilization, combining the similar historical fertilization case retrieval method, and making full use of the information of rice variety, planting plot and growth period, to recommend a refined fertilization plan including fertilization amount, fertilizer type and fertilization time for rice under different environmental characteristics and growth periods.

[0116] Step 3: Experimental analysis and evaluation

[0117] Taking the spatio-temporal data of rice fertilization crowd-sensing in a certain county as experimental data, constructing a rice fertilization knowledge graph that takes into account spatio-temporal characteristics. Through the visualization display of knowledge graph instances, describing the spatio-temporal evolution law and complex relationship of rice fertilization knowledge.

[0118] (2) Conducting model performance tests on 5 common rice nutrient disorders in a certain county, and evaluating the rice nutrient element diagnosis model using the accuracy evaluation index.

[0119] (3) Comparing the recommended results of the current rice fertilization plan in the experimental area with the experimental results of the model proposed in the present invention, including the comparison of three types of results: fertilization period, fertilization amount, and fertilizer type.

[0120] In the above steps, the construction of the rice fertilization knowledge graph considering spatio-temporal features, the rice nutrient element diagnosis model, and the fertilization plan recommendation model are the key points of the present invention, which will be discussed in detail below.

[0121] (1) Construction of the rice fertilization knowledge graph considering spatio-temporal features

[0122] (a) Ontological logical structure expression

[0123] Abstract the ontology in the field of rice fertilization into a six-tuple structure of concept, relationship, attribute, rule, instance, and state, and the formal representation is shown in Formula 1:

[0124] Onto = (Con, Rel, Prop, Rule, Ins, Sta) (Formula 1)

[0125] Among them, Con refers to the concept, which represents the general term of the set of things with the same characteristics in the fertilization knowledge system; Rel refers to the relationship, which represents the relationship types between rice fertilization concepts, between concepts and instances, and between instances; Prop refers to the attribute, which represents the semantic and spatio-temporal characteristics of different instance objects; Rule refers to the rule, which defines the constraint expression of the value range, type, and combination method of fertilization domain concepts and instances, so as to support semantic reasoning; Ins refers to the instance, which represents the specific expression based on the domain concept; Sta refers to the state, which represents the attribute characteristics of the fertilization instance under different spatio-temporal conditions.

[0126] (b) Concept system establishment

[0127] According to business procedures, scientific and technological literature, and existing ontology model systems, sort out the rice fertilization concepts, and establish a rice fertilization domain concept system including four core elements: rice, planting environment, nutrient element diagnosis, and fertilization plan, as Figure 1 shown.

[0128] (c) Instance-state-attribute representation

[0129] Guided by rice fertilization knowledge, define the instances, states, and attribute characteristics of rice fertilization concepts and their sub-concepts. The rice fertilization instance is the specific expression of the concept, and the description of the rice attribute characteristics is centered around the instance. At the same time, in order to reasonably represent the characteristics of fertilization knowledge in time or space, set state nodes to divide the periodic characteristics of the instance, Figure 2 showing the expression form of the fertilization plan in the knowledge graph.

[0130] (d) Relationship determination

[0131] Define the relationships in the field of rice fertilization as time relationship, space relationship, semantic relationship, and mathematical relationship, as Figure 3As shown in the figure. The time relationship of rice fertilization refers to the sequence of occurrence among rice fertilization examples, which is usually expressed by time topological relationships. The spatial relationship of rice fertilization aims to describe the spatial relationships among geographical entity objects in the planting environment. Since the geographical entity objects related to rice fertilization are usually face objects, such as administrative regions, farmland plots, etc., which are the scopes of rice production, the spatial orientation relationship and spatial distance relationship do not have obvious practical significance yet. Therefore, its spatial relationship is a spatial topological relationship, indicating the proximity and association degree among geographical entity objects. The semantic relationships of rice fertilization include inheritance relationships, instance relationships, and attribute relationships. Inheritance relationships represent the hierarchical structure between the upper and lower concepts of rice fertilization; instance relationships represent the relationships between concepts and instances; attribute relationships are the expressions of the mutual connections among entities at the semantic level and also the specific descriptions of the relationships between entities and attribute values. The mathematical relationships of rice fertilization represent the association relationships formed by calculating concepts or attributes through mathematical models.

[0132] (2) Rice Nutrient Element Diagnosis Model

[0133] As a physiological disease caused by nutrient deficiency or nutrient excess, nutrient disorders will cause obvious changes in the appearance characteristics of rice, such as morphology, growth vigor, leaf color, etc., showing typical symptoms. However, different types of nutrient disorders may present the same symptoms, and the diagnosis under different spatio-temporal conditions depends on the experience and knowledge of agricultural experts. Therefore, a single query statement cannot achieve the accurate diagnosis of rice nutrient disorders. In the classical Bayes' theorem, to diagnose the probability of a certain disease, a large number of conditional probability values are required, but it is difficult to accurately obtain conditional probability values in actual operations. And the reasoning based on the CF model can apply and process knowledge with uncertainty. Aiming at the problem that it is difficult to comprehensively judge the conclusion in knowledge reasoning when multiple rules support the conclusion, combining the CF model and the knowledge graph, a rice nutrient element diagnosis model is proposed. The specific process is as shown in the appendix Figure 4 As shown in the figure.

[0134] First, taking the appearance characteristics of rice and the planting plots in different planting areas as inputs, match the hazard symptoms and soil environmental elements with spatial characteristics in the spatio-temporal knowledge graph of rice fertilization. Secondly, based on the expert knowledge in the knowledge graph, further match the certainty factors of the hazard symptoms and the soil environment, and use formulas (2) - (3) to calculate the incidence probability of each nutrient disorder under the current rice planting situation. The nutrient disorder type corresponding to the highest incidence probability is taken as the diagnosis result.

[0135] CF(B,A) = CF(B,A1) + CF(B,A2) - CF(B,A1) × CF(B,A2) (Formula 2)

[0136]

[0137] Among them, A and B represent evidence and conclusion respectively. A1 and A2 are sub-evidences that make up evidence A. CF(B, A) represents the certainty factor, also known as the CF value. If CF(B, A) > 0, the larger the value of CF(B, A), the more the appearance of the evidence supports the conclusion to be true. If CF(B, A) < 0, the smaller the value of CF(B, A), the more the appearance of the evidence supports the conclusion to be false. If CF(B, A) = 0, the appearance of the evidence has no supporting effect on the conclusion. P(B|A) represents the diagnostic probability, and P(B) represents the prior probability of the conclusion, which can be set by expert experience or regarded as unknown, that is, P(B) = 0.5.

[0138] (3) Rice Fertilization Plan Recommendation Model

[0139] The fertilization plan recommendation model based on the knowledge graph can be divided into a problem description part and a solution part. The problem description part includes the rice growth period, rice variety, organic matter, available nitrogen, available phosphorus, available potassium, pH value, and yield level. The solution part includes the fertilization amount, fertilizer type, and fertilization time for the corresponding growth period of rice. Integrating the fertilization experience knowledge of multiple rice varieties and the principle of nutrient balance fertilization based on the rice fertilization knowledge graph, combined with the retrieval method of similar historical fertilization cases, a refined fertilization plan including fertilization amount, fertilizer type, and fertilization time is recommended for rice under different environmental characteristics and growth periods. The specific process is as Figure 5 shown.

[0140] First, input the information of rice variety, planting plot, and growth period. Match the problem description index information through the knowledge graph to obtain the spatio-temporal characteristics of the input information. Then use the method of knowledge representation learning to obtain the low-dimensional vector representation of all entities and relationships in the rice fertilization knowledge graph. On this basis, calculate the global similarity and retrieve similar historical fertilization cases. When the similarity of the retrieval result is relatively high, generate a rice fertilization plan. When there are no similar historical fertilization cases, combine the existing information in the graph and use the calculation method of the nutrient balance method to supplement the specific fertilization amount value. Finally, match the rice growth period with the fertilization plan to form the final rice fertilization plan.

[0141] (a) Retrieval of Similar Historical Fertilization Cases

[0142] The goal of case retrieval is to match historical cases with relatively high similarity in the problem description part of the case library according to the rice planting situation. According to the data type, the attributes of the problem description part can be divided into two categories: numerical attributes and text attributes.

[0143] Numerical attributes include organic matter, available nitrogen, available phosphorus, available potassium, pH value, and yield level. The similarity of the same numerical attribute between different cases can be represented by the distance after normalizing two values. The specific calculation process is as shown in Formula 4:

[0144]

[0145] In the formula and respectively represent the value of attribute k of the i-th case and the case j to be recommended, and respectively represent the maximum and minimum values of attribute k, represents and the numerical attribute similarity of

[0146] The text attributes are rice varieties and growth periods. When retrieving historical cases, it is necessary to calculate the similarity between different rice varieties. Since the similarity of text attributes cannot be directly measured by a computer, it is necessary to represent them in a vectorized form. Text attributes are stored in the form of entities and relationships in the knowledge graph, adopting a discrete and explicit symbolic representation. Some studies have shown that vector operations based on the vectorized representation of the knowledge graph can reflect the similarity between entities. In the rice fertilization knowledge graph, the attribute relationships are all one-to-one relationships. Therefore, the TransE model is used for knowledge representation learning. This model has a significant effect on modeling one-to-one relationships and can simply and efficiently capture inference features without manual design. The core idea of the TransE model is that when the triple (h, r, t) holds, it should satisfy formula 5, and the definition of the model scoring function is shown in formula 6:

[0147] h + r ≈ h (Formula 5)

[0148] f r (h, t) = ||h + t - r|| L1 / L2 (Formula 6)

[0149] Among them, h is the head entity, r is the relationship, t is the tail entity, and L1 and L2 are the first and second normal forms respectively. For a correct triple, it is expected that the smaller the score, the better; for an incorrect triple, it is expected that the higher the score, the better.

[0150] Using the distributed low-dimensional vectors of rice varieties obtained after knowledge graph representation learning as the numerical representation of various types of rice varieties, the cosine similarity between vectors is used to calculate the similarity between rice varieties. The cosine similarity calculation formula is as follows:

[0151]

[0152] Among them, and respectively represent the vectors of the rice variety attribute k of the case to be recommended and the i-th case, represents the cosine similarity between rice variety vectors.

[0153] Calculate the global similarity of cases using Formula 8, and use the historical cases with higher global similarity as the results of case retrieval:

[0154]

[0155] Among them, K represents the number of attributes, w k represents the weight of attribute k, s k represents the similarity of attribute k between two cases, and S i represents the global similarity between the i-th case and the case to be recommended. In this step, the weights of each attribute are set to be the same.

[0156] (b) Calculation by nutrient balance method

[0157] Since the rice planting situation is not completely the same as that of historical fertilization cases, it is necessary to correct the generated rice fertilization plan. The global similarity was obtained through the retrieval of similar historical fertilization cases above. When the global similarity of the retrieval results is less than the preset threshold, it is considered that there are no historical fertilization cases with high similarity. At this time, combined with the existing information in the knowledge graph, the calculation method of the nutrient balance method is used to supplement the specific fertilization amount value.

[0158] The nutrient balance method aims to achieve the balance of nutrient supply and demand between crops and soil, and calculates the amount of fertilizer required to achieve the target yield according to the difference between the amount of fertilizer required by the crops and the amount of fertilizer that the soil can provide. The calculation process of the recommended fertilization amount is shown in Formula 9:

[0159]

[0160] Among them, Fer refers to the required fertilization amount, Y represents the target yield, U represents the fertilizer utilization rate in the current season, W is the nutrient absorption amount per unit yield, C represents the nutrient content in the fertilizer, X represents the soil test value, and K represents the correction coefficient of soil available nutrients.

[0161] Taking a certain county as an example, the structured data used in the present invention is the cultivated land evaluation sample point database of a certain county, including soil basic data and spatial data and attribute data closely related to the process of nutrient precise management. The semi-structured data is the content of knowledge cards on the China Agricultural Information Network and encyclopedia websites, and the unstructured data is more than 300 rice fertilization-related articles published in journals such as "Scientia Agricultura Sinica", "Friends of Farmers' Prosperity", and "Southern Agriculture". In summary, more than 100,000 words of effective data are obtained, involving information such as rural cadastral surveys, agricultural censuses, crop fertilizer requirements, topography, and farmland water conservancy facilities.

[0162] Based on the rice fertilization knowledge graph construction method proposed above, the knowledge graph constructed in the experiment covers 1,768 nodes and contains 2,690 triples. Taking the fertilization plan "001" as an example, the representation result of the fertilization plan of the present invention is as Figure 6As shown. The results show that this method more completely demonstrates the status and attribute representation of the fertilization plan itself, as well as the semantic and spatio-temporal relationships between the fertilization plan and the rice and planting environment entities. Through the knowledge graph, agricultural knowledge such as fertilization plans at different growth stages can be efficiently associated with the rice planting situation, planting plots, and their spatial characteristics, providing support for rice nutrient element diagnosis and fertilization plan recommendation.

[0163] Based on the rice nutrient element diagnosis model proposed in the present invention, 5 common rice nutrient disorders in a certain county were diagnosed. A total of 50 samples were selected, and the accuracy evaluation index was used to evaluate the rice nutrient element diagnosis model. The diagnosis example based on the knowledge graph is as Figure 7 shown, and the experimental results are shown in Table 1. Calculated by formula (10), the accuracy rate of the rice nutrient element diagnosis method based on knowledge reasoning proposed in this paper is 84%, proving that this method can effectively diagnose the rice nutrient element situation.

[0164]

[0165] Among them, TP represents the number of cases where the model prediction result is consistent with the true result, and FP represents the number of cases where the model prediction result is inconsistent with the true result.

[0166] Table 1 Experimental results of rice nutrient element diagnosis

[0167]

[0168]

[0169] Based on the rice fertilization plan recommendation model proposed in the present invention, 15 experimental plots were randomly selected for verification. The recommended results of the current rice fertilization plan in the experimental area were compared with the experimental results of the model proposed in the present invention, including the comparison of three types of results: fertilization period, fertilization amount, and fertilizer type. At present, the main reference basis for farmers in a certain county to make fertilization decisions is the local current rice crop fertilization guidance manual, and its recommended results are shown in Table 2. The fertilization plan recommendation example based on the knowledge graph of the rice fertilization plan recommendation model proposed in the present invention is as Figure 8 shown. By means of a questionnaire survey, the proposed method in this paper was compared with the recommended plan of the guidance manual. The total number of survey objects was 6, and all of them were experts in the field of rice fertilization in a certain county and had the ability of independent judgment and fertilization decision-making. The experimental results are as Figure 9 shown, and the results show that the average score of the recommended plan of the model in this paper is higher than that of the recommended plan of the guidance manual, proving that the fertilization plan recommendation method in this paper can provide more applicable fertilization plan suggestions in actual production.

[0170] Table 2 Example of recommended results of the current fertilization guidance manual in a certain county

[0171]

[0172] The above are the preferred embodiments of the present invention. All changes made according to the technical solution of the present invention, when the functions and effects produced do not exceed the scope of the technical solution of the present invention, shall fall within the protection scope of the present invention.

Claims

1. A method for constructing a rice fertilization knowledge graph and knowledge reasoning that takes into account spatio-temporal characteristics, characterized in that, it includes: S1. Construct a rice fertilization knowledge graph that takes into account spatio-temporal characteristics: Design the ontology of the rice fertilization knowledge graph, complete the knowledge modeling of the rice fertilization knowledge graph that takes into account spatio-temporal characteristics, and on this basis, perform rice fertilization knowledge extraction and knowledge storage; S2. Propose a rice fertilization strategy reasoning method based on the knowledge graph, including the construction of a rice nutrient element diagnosis model and a rice fertilization plan recommendation model; S3. Use the spatio-temporal data of rice fertilization crowd-sensing as experimental data, and analyze and evaluate the rice fertilization strategy reasoning results based on the constructed knowledge graph; The specific implementation of step S1 is as follows: (1) Knowledge modeling of the rice fertilization knowledge graph that takes into account spatio-temporal characteristics: Using a top-down approach, through ontology logical structure expression, concept system establishment, instance-state-attribute representation, and relationship determination, complete the knowledge modeling of the rice fertilization knowledge graph; (2) Rice fertilization knowledge extraction and knowledge storage: For structured data, form a mapping from data tables to RDF according to the rice fertilization pattern ontology, and directly convert it into a knowledge graph expression; for semi-structured and unstructured data, first perform preprocessing operations on the data to obtain valid data, then perform template-based automated information extraction, and finally organize and correct the obtained information; obtain time relationships through discrimination and sorting of the extracted time elements, and use the nine-intersection model DE-9IM based on dimension expansion to describe the spatial relationships between geographical entity objects, as shown in the following formula: wherein, R DE-9IM(a,b) represents the spatial topological relationship between the geographical entities described by DE-9IM; a and b represent different rice geographical entities; I, B, and E respectively represent the interior, boundary, and exterior of the geographical entity; the dim function returns the maximum dimension of the intersecting geometric bodies; Use the Neo4j graph database. Based on the Resource Description Framework, represent various relevant elements of rice fertilization knowledge in the form of (entity, relationship, entity) and (entity, attribute, attribute value) triples. Subsequently, use the graph database Neo4j to store rice fertilization knowledge in a "node-relationship" storage model. The graph database stores the entities and attribute values in the triples as nodes, and stores the attributes and relationships as edges; The specific implementation method of the knowledge modeling of the rice fertilization knowledge graph that takes into account spatio-temporal characteristics is as follows: (a) Ontology logical structure expression of the rice fertilization knowledge graph Abstract the ontology of the rice fertilization field into a six-tuple structure of concept, relationship, attribute, rule, instance, and state, and the formal representation is as shown in the following formula: Onto = (Con, Rel, Prop, Rule, Ins, Sta) Among them, Con refers to the concept, which represents the general collection of things with the same characteristics in the rice fertilization knowledge system; Rel refers to the relationship, which represents the relationship types between rice fertilization concepts, between concepts and instances, and between instances; Prop refers to the attribute, which represents the semantic and spatio-temporal characteristics of different instance objects; Rule refers to the rule, which defines the constraint expressions of the value range, type, and combination method of rice fertilization field concepts and instances, so as to support semantic reasoning; Ins refers to the instance, which represents the specific expression based on the domain concept; Sta refers to the state, which represents the attribute characteristics of rice fertilization instances under different spatio-temporal conditions; (b) Concept system establishment Based on the rice fertilization knowledge graph ontology model, the concept of rice fertilization was sorted out, and a concept system of rice fertilization was established, including four core elements: rice, planting environment, nutrient element diagnosis, and fertilization plan. (c) Instance-State-Attribute Representation Guided by rice fertilization knowledge, the instance, state and attribute characteristics of the rice fertilization concept and its sub-concepts are defined; the rice fertilization instance is a concrete expression of the concept, and the description of rice attribute characteristics revolves around the instance. At the same time, in order to express the characteristics of fertilization knowledge in time or space, the periodic characteristics of the state node division instance are set; (d) Relationship determination The relationships in the field of rice fertilization are defined as temporal relationships, spatial relationships, semantic relationships, and mathematical relationships. The temporal relationship of rice fertilization refers to the order in which rice fertilization instances occur, which is expressed by temporal topological relationships. The spatial relationship of rice fertilization aims to describe the spatial relationship between geographic entity objects in the planting environment, and its spatial relationship is a spatial topological relationship, which indicates the proximity and association degree between geographic entity objects. The semantic relationship of rice fertilization includes inheritance relationship, instance relationship, and attribute relationship. The inheritance relationship indicates the hierarchical structure between the upper and lower concepts of rice fertilization. The instance relationship indicates the relationship between concepts and instances. The attribute relationship is the expression of the mutual connection between entities at the semantic level, and is also a specific description of the relationship between entities and attribute values. The mathematical relationship of rice fertilization indicates the association relationship between concepts or attributes calculated by mathematical models. Step S2 is specifically implemented as follows: (1) Constructing a rice nutrient element diagnostic model: combining the deterministic factor CF model with the knowledge graph, making full use of the rice appearance representation and planting plots in different planting areas, matching the damage symptoms in the rice fertilization spatiotemporal knowledge graph and the soil environmental elements with spatial characteristics, obtaining the CF value that can support multiple rule diagnosis conclusions, and calculating the types of nutritional disorders that rice may suffer from and their probabilities; (2) Construct a rice fertilization scheme recommendation model: Based on the rice fertilization knowledge graph, the fertilization experience knowledge of multiple rice varieties and the principle of nutrient balance fertilization are integrated, combined with the similar historical fertilization case retrieval method, and the rice variety, planting plot and growth period information are fully utilized to recommend refined fertilization schemes including fertilizer amount, fertilizer type and fertilization time for rice under different environmental characteristics and growth periods; The specific implementation method of constructing a rice nutrient element diagnostic model is as follows: First, the rice appearance and planting plots in different planting areas are used as input to match the damage symptoms and soil environmental elements with spatial characteristics in the rice fertilization knowledge map; secondly, based on the expert knowledge of the knowledge map, the damage symptoms and the deterministic factors of the soil environment are further matched, and the probability of each nutritional disorder under the current rice planting conditions is calculated using the following two formulas, and the type of nutritional disorder corresponding to the highest probability of occurrence is used as the diagnosis result; CF(B,A)=CF(B,A1)+CF(B,A2)-CF(B,A1)×CF(B,A2) Among them, A and B represent evidence and conclusion respectively, A1 and A2 are sub-evidences that make up evidence A, and CF(B, A) represents the certainty factor, also known as the CF value. If CF(B, A) > 0, the larger the value of CF(B, A), the more the appearance of the evidence supports the conclusion to be true. If CF(B, A) < 0, the smaller the value of CF(B, A), the more the appearance of the evidence supports the conclusion to be false. If CF(B, A) = 0, the appearance of the evidence has no supporting effect on the conclusion. P(B|A) represents the diagnostic probability, and P(B) represents the prior probability of the conclusion. The specific implementation method for constructing a rice fertilization plan recommendation model is as follows: First, input information on rice varieties, planting plots, and growth periods, match the problem description index information through the rice fertilization knowledge graph, and obtain the spatio-temporal characteristics of the input information. Then, use the method of knowledge representation learning to obtain the low-dimensional vector representations of all entities and relationships in the rice fertilization knowledge graph. On this basis, calculate the global similarity and retrieve similar historical fertilization cases. When the similarity of the retrieval result is greater than or equal to the preset threshold, generate a rice fertilization plan. When there are no similar historical fertilization cases, combine the existing information in the rice fertilization knowledge graph and use the calculation method of the nutrient balance method to supplement the specific fertilization amount value. Finally, match the rice growth period with the fertilization plan to form the final rice fertilization plan. (a) Retrieval of similar historical fertilization cases The goal of retrieving similar historical fertilization cases is to match historical cases in the case library with a similarity greater than or equal to the preset threshold in the problem description part according to the rice planting situation. According to the data type, the attributes of the problem description part are divided into two categories: numerical attributes and text attributes. Numerical attributes include organic matter, available nitrogen, available phosphorus, available potassium, pH value, and yield level. The similarity of the same numerical attribute between different cases is represented by the distance after normalizing two values. The specific calculation process is shown in the following formula: where and represent the values of the k-th attribute of the i-th case and the case j to be recommended, respectively, and represent the maximum and minimum values of the k-th attribute, respectively, represents the numerical attribute similarity between Text attributes are rice varieties and growth periods. When retrieving similar historical fertilization cases, it is necessary to calculate the similarity between different rice varieties. Since the similarity of text attributes cannot be directly measured by a computer, it is necessary to vectorize it. Text attributes are stored in the form of entities and relationships in the knowledge graph and are represented in a discrete and explicit symbolic form. Vector operations based on the vector representation of the knowledge graph can reflect the similarity between entities. The attribute relationships in the rice fertilization knowledge graph are all one-to-one relationships. Therefore, the TransE model is used for knowledge representation learning. The core idea of the TransE model is that when the triple (h, r, t) holds, it should satisfy the following formula: h + r ≈ t The definition of the model scoring function is shown in the following formula: f r (h, t) = ||h + t - r|| L1 / L2 Among them, h is the head entity, r is the relationship, t is the tail entity, and L1 and L2 are the first and second normal forms respectively. For a correct triple, it is expected that its score is as small as possible. For an incorrect triple, it is expected that its score is as high as possible. Taking the distributed low-dimensional vectors of rice varieties obtained after knowledge graph representation learning as the numerical representation of various types of rice varieties, the cosine similarity between vectors is used to calculate the similarity between rice varieties. The cosine similarity calculation formula is as follows: Among them, and represent the vectors of the rice variety attribute k of the case to be recommended and the i-th case respectively, represents the cosine similarity between the rice variety vectors; The global similarity of the case is calculated using the following formula, and the historical cases with global similarity greater than or equal to the preset threshold are used as the results of case retrieval: Among them, K represents the number of attributes, w k represents the weight of attribute k, s k represents the similarity of attribute k between two cases, S i represents the global similarity between the i-th case and the case to be recommended; (b) Nutrient balance method calculation The nutrient balance method aims to achieve the supply-demand balance of nutrients between crops and soil. The amount of fertilizer required to achieve the target yield is calculated based on the difference between the amount of fertilizer required by the crop and the amount of fertilizer that the soil can provide. The calculation process of the recommended fertilization amount is shown in the following formula: Among them, Fer refers to the amount of fertilizer required, Y represents the target yield, U represents the fertilizer utilization rate in the current season, W is the nutrient absorption amount per unit yield, C represents the nutrient content in the fertilizer, X represents the soil test value, and K represents the correction coefficient of soil available nutrients.

2. The method for constructing a rice fertilization knowledge graph and knowledge reasoning considering spatio-temporal characteristics according to claim 1, characterized in that Step S3 is specifically implemented as follows: (1) Perform model performance tests on 5 common rice nutrient disorders, and evaluate the rice nutrient element diagnosis model using the accuracy evaluation index; (2) Compare the recommended results of the current rice fertilization plan in the experimental area with the experimental results of the rice fertilization plan recommendation model, including the comparison of three types of results: fertilization period, fertilization amount, and fertilizer type.

Images