A method and system for constructing a smart orchard expert knowledge base based on machine learning

Abstract

The present invention provides a method and system for constructing a smart orchard expert knowledge base based on machine learning. The method comprises the following steps: collecting and preprocessing orchard-related data to obtain an initial knowledge base; constructing a knowledge representation model based on the initial knowledge base using an ontology construction method, and obtaining a standardized knowledge representation based on the knowledge representation model; obtaining a knowledge reasoning chain based on the standardized knowledge representation using machine learning reasoning; using an incremental learning algorithm based on the knowledge reasoning chain to learn new planting cases and expert feedback, automatically updating the initial knowledge base to obtain an expert knowledge base; and using a machine learning algorithm based on the expert knowledge base to obtain orchard management decisions, thereby completing the construction of a smart orchard expert knowledge base based on machine learning. This invention can significantly improve the intelligent level of orchard management and provide strong support for precision planting.

Description

Technical Field

[0001] The present invention belongs to the field of machine learning technology, and specifically relates to a method and system for constructing a smart orchard expert knowledge base based on machine learning. Background Art

[0002] The development of a smart orchard requires systematically and structured organization of the experiential knowledge of fruit tree planting experts to form a knowledge base that can be accessed by intelligent decision-making systems. However, this construction process presents a series of technical challenges, including knowledge acquisition, representation, and reasoning. First, fruit tree planting involves knowledge from multiple domains, including the environment, pest and disease control, and cultivation management. This knowledge comes from diverse sources and in various forms, lacking a unified collection and representation method. Second, fruit tree growth cycles are long and influenced by complex and changing factors. The experiential knowledge of planting experts is often subjective and uncertain, making it difficult to directly transform into a structured and computable form. Furthermore, different knowledge bases may be interconnected and constrained, requiring the establishment of a sound knowledge organization and reasoning mechanism to effectively apply this knowledge. Therefore, the construction of a smart orchard knowledge base requires not only agricultural domain knowledge but also artificial intelligence technologies such as machine learning and knowledge engineering. By bridging the gap between agricultural knowledge and information technology, we can mine implicit knowledge and establish a coherent, comprehensive, and interpretable knowledge base, providing a solid foundation for intelligent decision-making in smart orchards. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the present invention provides a method and system for constructing a smart orchard expert knowledge base based on machine learning, which completes the smart management of the orchard by constructing an expert knowledge base that is updated in a timely manner.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A method for constructing a smart orchard expert knowledge base based on machine learning includes the following steps:

[0006] Collect and preprocess orchard-related data to obtain an initial knowledge base;

[0007] Based on the initial knowledge base, a knowledge representation model is constructed using an ontology construction method, and a standardized knowledge representation is obtained based on the knowledge representation model;

[0008] Based on the standardized knowledge representation, machine learning reasoning is used to obtain a knowledge reasoning chain;

[0009] Based on the knowledge reasoning chain, an incremental learning algorithm is used to learn new planting cases and expert feedback, and the initial knowledge base is automatically updated to obtain an expert knowledge base;

[0010] Based on the expert knowledge base, a machine learning algorithm is used to obtain orchard management decisions, completing the construction of a smart orchard expert knowledge base based on machine learning.

[0011] Preferably, the orchard-related data include orchard meteorological data, soil temperature and humidity data, water and fertilizer application amounts, pest and disease control data, fruit tree growth cycles, and historical yields.

[0012] Preferably, the method for constructing the knowledge representation model is:

[0013] Based on natural language processing technology, the initial knowledge base is subjected to knowledge segmentation, part-of-speech tagging, and entity recognition to obtain knowledge keywords;

[0014] Classifying the knowledge keywords based on a clustering algorithm to obtain entities, and extracting attributes and relationships of each entity;

[0015] Based on the entities, the attributes and the relationships, an ontology construction method is adopted to obtain a knowledge formalization representation model;

[0016] Based on the knowledge formalization representation model, a graph embedding algorithm is used to learn low-dimensional vector representations of the entities and the relationships to obtain semantic representations of the knowledge keywords;

[0017] The knowledge representation model is obtained based on the semantic representation and the knowledge formalization representation model.

[0018] Preferably, the method for obtaining the knowledge reasoning chain is:

[0019] Based on the convolutional information propagation network, the relationship between the head entity and the tail entity is updated to obtain the contextual features of the entity;

[0020] Based on the random walk algorithm, the relationship paths between adjacent entities are mined to obtain the set of all relationship paths between entity pairs;

[0021] Performing semantic feature encoding on the relationship path set based on a bidirectional gated recurrent unit network to obtain semantic vectors of all relationship paths;

[0022] Aggregate the semantic vectors of different relationship paths based on the contextual features of the entity to obtain the relationship path features of the entity pair;

[0023] The context features and the relationship path features are fused and added, and the knowledge reasoning chain is obtained using a softmax function.

[0024] Preferably, the method for obtaining the contextual features of an entity is:

[0025] Based on average pooling, aggregate the attribute information of adjacent entities of each entity to obtain aggregated neighbor information;

[0026] Use the convolution kernel to convolve the aggregated neighbor information of each entity and update the entity representation;

[0027] Based on the updated entity representation, convolutional information propagation is completed;

[0028] During the convolutional information propagation process, the head entity and the tail entity exchange and update information through their respective neighbor entities, and update the relationship representation between the entities;

[0029] Based on the update of the entity representation and the relationship representation, each entity obtains an embedding vector containing local graph structure information;

[0030] Based on the embedding vector, the context feature of the entity is obtained.

[0031] Preferably, the method for obtaining the expert knowledge base is:

[0032] Preprocess and extract features of new implant case data and expert feedback data to obtain new data features;

[0033] Identifying the new data features based on the knowledge reasoning chain, and screening out new data features whose similarity with existing data in the initial knowledge base does not meet a preset threshold;

[0034] Labels are added to the screened new data features as incremental inputs, and the new planting case data and expert feedback data are integrated into the initial knowledge base in batches through a small batch learning algorithm to obtain the expert knowledge base.

[0035] Preferably, the method for obtaining orchard management decisions is:

[0036] For the user's decision-making problems, based on the historical data in the expert knowledge base, a decision tree algorithm is used to classify the knowledge to obtain knowledge sets of different categories;

[0037] Based on different categories of knowledge sets, the support vector machine algorithm is used to establish a knowledge prediction model to obtain prediction results;

[0038] The prediction results are optimized based on a genetic algorithm to obtain intelligent decision-making suggestions for orchard management.

[0039] The present invention also provides a system for constructing a smart orchard expert knowledge base based on machine learning, which is used to implement the method, including:

[0040] Data acquisition module, used to collect and pre-process orchard-related data to obtain an initial knowledge base;

[0041] A knowledge standardization module, configured to construct a knowledge representation model based on the initial knowledge base using an ontology construction method, and obtain standardized knowledge representation based on the knowledge representation model;

[0042] A knowledge reasoning chain acquisition module, configured to obtain a knowledge reasoning chain based on the standardized knowledge representation and using machine learning reasoning;

[0043] An incremental module is used to learn new planting cases and expert feedback based on the knowledge reasoning chain using an incremental learning algorithm, automatically update the initial knowledge base, and obtain an expert knowledge base;

[0044] The intelligent decision-making module is used to obtain orchard management decisions based on the expert knowledge base and adopt a machine learning algorithm to complete the construction of a smart orchard expert knowledge base based on machine learning.

[0045] Compared with the existing technology, the beneficial effects of the present invention are as follows: the intelligent orchard expert knowledge base based on machine learning can extract useful information from a large amount of orchard data through automated data collection and processing, and update the knowledge base in real time to ensure the accuracy and timeliness of the knowledge. Historical data and real-time data are used for intelligent reasoning and decision support, providing orchard managers with more accurate and reliable decision-making basis. It has good domain adaptability and scalability, and can be easily expanded to other orchards or crop fields to meet the needs of different orchard managers. The present invention realizes the transformation from scattered original knowledge to structured knowledge base, and then to the whole process of intelligent application, solving the problems of knowledge fragmentation, update lag, application limitations, etc. in traditional orchard management. Based on the standardized representation, dynamic evolution and intelligent application of knowledge, the present invention can significantly improve the intelligence level of orchard management and provide strong support for precision planting. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 This is a flow chart of a method for constructing a smart orchard expert knowledge base based on machine learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0049] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] Example 1

[0051] like Figure 1 As shown, a method for constructing a smart orchard expert knowledge base based on machine learning includes the following steps:

[0052] S1: Collect and pre-process orchard-related data to obtain an initial knowledge base; a further implementation method is that the orchard-related data includes orchard meteorological data, soil temperature and humidity data, water and fertilizer application amount, pest and disease control data, fruit tree growth cycle and historical yields.

[0053] S2: Based on the initial knowledge base, the ontology construction method is used to build a knowledge representation model, and standardized knowledge representation is obtained based on the knowledge representation model.

[0054] A further implementation method is to construct a knowledge representation model by:

[0055] Based on natural language processing technology, the initial knowledge base is segmented, POS-tagged, and entity recognized to obtain knowledge keywords;

[0056] Classify knowledge keywords based on clustering algorithms, obtain entities, and extract the attributes and relationships of each entity;

[0057] Based on entities, attributes and relationships, ontology construction method is adopted to obtain the formal representation model of knowledge;

[0058] Based on the knowledge formalization model, a graph embedding algorithm is used to learn low-dimensional vector representations of entities and relationships to obtain semantic representations of knowledge keywords.

[0059] Based on semantic representation and knowledge formal representation model, a knowledge representation model is obtained.

[0060] In the present embodiment, after building the initial knowledge base, the knowledge text is first segmented using the jieba word segmentation tool, and then the THULAC tool is used for part-of-speech tagging, and the key concepts such as crop names, pest and disease names, and farming operations in the text are identified by the BiLSTM-CRF model. Then, the extracted keywords are clustered into different categories such as environmental monitoring, pest control, and cultivation management according to semantic similarity using the K-means clustering algorithm. For each category, the core attributes of the category are extracted through word frequency statistics and TF-IDF weight calculation, such as environmental monitoring including temperature, humidity, light, etc., and the correlation between attributes is analyzed, such as temperature and humidity are negatively correlated. On this basis, the ontology construction tool Protégé is used to define the ontology model of agricultural knowledge, including concept class (entity), attribute class and relationship class, to form a knowledge formalization representation model.

[0061] Knowledge from diverse sources, such as agricultural forums and meteorological websites, was mapped into a unified ontology model through knowledge extraction and entity alignment techniques. Finally, the Neo4j graph database was used to store the agricultural knowledge graph. The entities included 4,326 concepts, such as crops, pests and diseases, and agricultural operations, and the edges included 6,851 attribute associations and causal relationships. To mine the implicit semantics between knowledge, knowledge representation learning algorithms such as TransE were used to learn a 64-dimensional vector representation for each node (entity) and edge (relationship). Semantic relevance between concepts was calculated using cosine similarity, and the knowledge representation model was constructed by integrating explicit and implicit semantics.

[0062] S3: Based on the standardized knowledge representation, machine learning reasoning is used to obtain a knowledge reasoning chain. A further implementation method is to obtain the knowledge reasoning chain by:

[0063] S31: Based on the convolutional information propagation network, the relationship between the head entity and the tail entity is updated to obtain the contextual features of the entity.

[0064] A further implementation method is that the method for obtaining the context feature of the entity is:

[0065] S311: Based on average pooling, aggregate the attribute information of the adjacent entities of each entity to obtain aggregated neighbor information;

[0066] S312: Convolve the aggregated neighbor information of each entity using the convolution kernel to update the entity representation; complete the convolution information propagation based on the updated entity representation;

[0067] S313: During the convolutional information propagation process, the head entity and the tail entity exchange and update information through their respective neighbor entities, and update the relationship representation between the entities;

[0068] S314: Based on the update of entity representation and relationship representation, each entity obtains an embedding vector containing local graph structure information;

[0069] S315: Based on the embedding vector, obtain the context feature of the entity.

[0070] S32: Based on the random walk algorithm, the relationship paths between adjacent entities are mined to obtain the set of all relationship paths between entity pairs;

[0071] Randomly select the initial entity of the walk and obtain the transition probability from the current entity to the adjacent entity;

[0072] The termination condition of the walk is to visit the preset termination entity. During the walk, the path passed is recorded and the target entity in the path is marked.

[0073] In a single walk, starting from the starting entity, the next entity is randomly selected according to the transition probability until the termination condition of the walk is met. In each step, it is checked whether the terminal entity is reached, and all paths containing the terminal entity are recorded.

[0074] Among all the collected paths, there may be duplicate paths. Duplicate paths are removed by comparing the node sequences in the paths. All non-duplicate paths are merged into a set, which is the set of all relationship paths between the starting entity and the ending entity.

[0075] This embodiment also includes setting a maximum path length limit to reduce the amount of calculation, and evaluating the path value based on the importance and frequency of the relationships in the path to optimize the random walk algorithm.

[0076] S33: Based on the bidirectional gated recurrent unit network (BiGRU), the semantic features of the relationship path set are encoded to obtain the semantic vectors of all relationship paths.

[0077] The forward GRU and backward GRU are used to encode the relation path respectively, and the forward GRU hidden state and the backward GRU hidden state are obtained. The two hidden states are concatenated, and then the concatenated hidden states are linearly transformed to obtain the semantic vector of the relation path after GRU encoding.

[0078] S34: Aggregate the semantic vectors of different relationship paths based on the contextual features of the entity to obtain the relationship path features of the entity pair; in this embodiment, calculate the weight of each relationship path based on the contextual features of the entity, perform weighted addition on all relationship paths, and obtain the relationship path features.

[0079] S35: The context features and relationship path features are fused and added together, and the softmax function is used to obtain the knowledge reasoning chain.

[0080] S4: Based on the knowledge reasoning chain, an incremental learning algorithm is used to learn new planting cases and expert feedback, automatically update the initial knowledge base, and obtain an expert knowledge base; a further implementation method is to obtain the expert knowledge base by:

[0081] Preprocess and extract features of new implant case data and expert feedback data to obtain new data features;

[0082] Identify new data features based on the knowledge reasoning chain and filter out new data features whose similarity with existing data in the initial knowledge base does not meet the preset threshold;

[0083] Labels are added to the filtered new data features as incremental inputs. Through the small batch learning algorithm, the new planting case data and expert feedback data are integrated into the initial knowledge base in batches to obtain the expert knowledge base.

[0084] Specifically, when obtaining planting cases and expert feedback data, web crawler technology is used to capture data from network resources such as agricultural forums and planting blogs. The acquired data is processed (the same as the processing process of steps S2, S3, and S4), and natural language processing technologies such as word segmentation, part-of-speech tagging, and named entity recognition are used to extract key information such as crop name, planting time, fertilizer application amount, and pest and disease control measures. Based on the extracted key information, an ontological knowledge representation model is constructed, and the RDF (Resource Description Framework) format is used to organize the knowledge into subject-predicate-object triples. For the knowledge representation model, the newly acquired planting cases and expert feedback data are incrementally integrated into the existing knowledge base. By setting the knowledge decay factor, outdated knowledge is forgotten, and the knowledge freshness index is introduced to quantify the timeliness of knowledge. At the same time, online learning algorithms are used to monitor data flows in real time and capture new knowledge patterns. When the support of a knowledge pattern exceeds a threshold (such as 8), the update of the knowledge base is triggered (i.e., new data features are identified based on the knowledge reasoning chain, and new data features whose similarity with existing data in the initial knowledge base does not meet the preset threshold are screened).

[0085] The KGEM algorithm (Knowledge Graph Embedding Model) is used to learn the low-dimensional vector representation of knowledge in the expert knowledge base. Through vector operations, the evolution direction of knowledge is predicted. Based on the timestamp information of knowledge, a knowledge evolution time series is constructed. Using the time series prediction model, future knowledge development trends are predicted, realizing the dynamic evolution of the knowledge base.

[0086] S5: Based on the expert knowledge base, machine learning algorithms are used to obtain orchard management decisions and complete the construction of a smart orchard expert knowledge base based on machine learning.

[0087] In this embodiment, the reasoning process of the smart orchard expert knowledge base is as follows:

[0088] 1. Input orchard data (including environmental parameters, crop growth data, etc.);

[0089] 2. Analyze and predict through machine learning models;

[0090] 3. Output decision-making recommendations (such as irrigation recommendations, fertilization recommendations, etc.).

[0091] A further embodiment is that the method for obtaining orchard management decisions is:

[0092] For user decision-making problems, we use the decision tree algorithm to classify knowledge based on historical data in the expert knowledge base to obtain knowledge sets of different categories.

[0093] Based on different categories of knowledge sets, the support vector machine algorithm is used to establish a knowledge prediction model to obtain prediction results;

[0094] The prediction results are optimized based on genetic algorithms to obtain intelligent decision-making suggestions for orchard management.

[0095] Specifically, the C5 decision tree algorithm was used to classify knowledge based on historical data on environmental parameters and growth conditions over the past five years from the Smart Orchard Expert Knowledge Base. By calculating the information gain ratio of each attribute, five attributes, including temperature, humidity, and light, were selected as the classification basis, and the knowledge was divided into eight categories, including irrigation, fertilization, and pest control. A support vector machine algorithm was used to build a prediction model for each knowledge set. Taking irrigation knowledge as an example, soil moisture, air temperature, and evaporation were selected as input features, and the irrigation volume for the next seven days was used as the output. Using a Gaussian kernel function, the model parameters were optimized through grid search, achieving a prediction accuracy of 95% on the test set. Furthermore, a genetic algorithm was used to optimize the prediction results, with water savings and yield as the optimization objectives. A fitness function was designed, and the optimal irrigation decision plan was obtained through 500 generations of selection, crossover, and mutation operations. This solution was applied to the orchard's automatic irrigation system, dynamically adjusting irrigation duration and water volume based on real-time data from soil moisture sensors. At the same time, various parameters of the irrigation process are continuously collected, the knowledge base is updated weekly, and the prediction model is incrementally trained using the gradient descent method to improve the model's generalization ability. When encountering extreme weather conditions such as three consecutive days of high temperature and drought, the system automatically activates the emergency plan. Based on the data analysis results in the knowledge base, it increases the frequency of irrigation. In combination with plant physiology models, it predicts the impact of high temperatures on fruit tree growth and takes cooling measures such as shading and foliar spraying in advance. Through data visualization technology, irrigation decisions and real-time monitoring data are displayed on the management platform in the form of line graphs and heat maps, and daily, weekly, and monthly irrigation reports are generated to assist orchard managers in analysis and decision-making. The system evaluates irrigation effectiveness based on indicators such as yield and water and fertilizer utilization, forming a closed-loop feedback loop and continuously optimizing irrigation strategies and the expert knowledge base.

[0096] Example 2

[0097] The present invention also provides a system for constructing a smart orchard expert knowledge base based on machine learning, which is used to implement the method, including:

[0098] Data acquisition module, used to collect and pre-process orchard-related data to obtain an initial knowledge base;

[0099] The knowledge standardization module is used to construct a knowledge representation model based on the initial knowledge base using an ontology construction method, and obtain standardized knowledge representation based on the knowledge representation model;

[0100] A knowledge reasoning chain acquisition module is used to obtain a knowledge reasoning chain based on standardized knowledge representation and machine learning reasoning;

[0101] The incremental module is used to learn new implant cases and expert feedback based on the knowledge reasoning chain using an incremental learning algorithm, automatically updating the initial knowledge base and obtaining the expert knowledge base;

[0102] The intelligent decision-making module is used to obtain orchard management decisions based on the expert knowledge base and adopt machine learning algorithms to complete the construction of the smart orchard expert knowledge base based on machine learning.

[0103] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A method for constructing a smart orchard expert knowledge base based on machine learning, characterized in that: The following steps are involved: Collect and preprocess orchard-related data to obtain an initial knowledge base; Based on the initial knowledge base, a knowledge representation model is constructed using an ontology construction method, and a standardized knowledge representation is obtained based on the knowledge representation model; Based on the standardized knowledge representation, machine learning reasoning is used to obtain a knowledge reasoning chain; Based on the knowledge reasoning chain, an incremental learning algorithm is used to learn new planting cases and expert feedback, and the initial knowledge base is automatically updated to obtain an expert knowledge base; Based on the expert knowledge base, a machine learning algorithm is used to obtain orchard management decisions, thereby completing the construction of a smart orchard expert knowledge base based on machine learning; The orchard-related data include orchard meteorological data, soil temperature and humidity data, water and fertilizer application, pest and disease control data, fruit tree growth cycle and historical yields; The method to obtain the expert knowledge base is: Preprocess and extract features of new implant case data and expert feedback data to obtain new data features; Identifying the new data features based on the knowledge reasoning chain, and screening out new data features whose similarity with existing data in the initial knowledge base does not meet a preset threshold; Label the filtered new data features as incremental input, and integrate the new planting case data and expert feedback data into the initial knowledge base in batches through a small batch learning algorithm to obtain the expert knowledge base; When acquiring planting cases and expert feedback data, web crawler technology is used to capture data from agricultural forums and planting blog network resources. The acquired data is processed using natural language processing technology, including word segmentation, part-of-speech tagging, and named entity recognition, to extract key information, including crop names, planting time, fertilizer application amount, and pest and disease control measures. Based on the extracted key information, an ontological knowledge representation model is constructed, using the RDF format to organize the knowledge into subject-predicate-object triples. For the knowledge representation model, newly acquired planting cases and expert feedback data are incrementally integrated into the existing knowledge base. By setting a knowledge decay factor, outdated knowledge is forgotten, and a knowledge freshness index is introduced to quantify the timeliness of knowledge. At the same time, an online learning algorithm is used to monitor data streams in real time and capture new knowledge patterns. When the support of a knowledge pattern exceeds a threshold, an update of the knowledge base is triggered. The KGEM algorithm is used to learn the low-dimensional vector representation of knowledge in the expert knowledge base and predict the evolution direction of knowledge through vector operations. Based on the timestamp information of knowledge, a knowledge evolution time series is constructed, and the time series prediction model is used to predict the development trend of future knowledge and realize the dynamic evolution of the knowledge base.

2. The method for constructing a smart orchard expert knowledge base based on machine learning according to claim 1, wherein: The method of constructing the knowledge representation model is: Based on natural language processing technology, the initial knowledge base is subjected to knowledge segmentation, part-of-speech tagging, and entity recognition to obtain knowledge keywords; Classifying the knowledge keywords based on a clustering algorithm to obtain entities, and extracting attributes and relationships of each entity; Based on the entities, the attributes and the relationships, an ontology construction method is adopted to obtain a knowledge formalization representation model; Based on the knowledge formalization representation model, a graph embedding algorithm is used to learn low-dimensional vector representations of the entities and the relationships to obtain semantic representations of the knowledge keywords; The knowledge representation model is obtained based on the semantic representation and the knowledge formalization representation model.

3. The method for constructing a smart orchard expert knowledge base based on machine learning according to claim 1, wherein: The method for obtaining the knowledge reasoning chain is: Based on the convolutional information propagation network, the relationship between the head entity and the tail entity is updated to obtain the contextual features of the entity; Based on the random walk algorithm, the relationship paths between adjacent entities are mined to obtain the set of all relationship paths between entity pairs; Performing semantic feature encoding on the relationship path set based on a bidirectional gated recurrent unit network to obtain semantic vectors of all relationship paths; Aggregate the semantic vectors of different relationship paths based on the contextual features of the entity to obtain the relationship path features of the entity pair; The context features and the relationship path features are fused and added, and the knowledge reasoning chain is obtained using a softmax function.

4. The method for constructing a smart orchard expert knowledge base based on machine learning according to claim 3, wherein: The method to obtain the context features of an entity is: Based on average pooling, aggregate the attribute information of adjacent entities of each entity to obtain aggregated neighbor information; Use the convolution kernel to convolve the aggregated neighbor information of each entity and update the entity representation; Based on the updated entity representation, convolutional information propagation is completed; During the convolutional information propagation process, the head entity and the tail entity exchange and update information through their respective neighbor entities, and update the relationship representation between the entities; Based on the update of the entity representation and the relationship representation, each entity obtains an embedding vector containing local graph structure information; Based on the embedding vector, the context feature of the entity is obtained.

5. The method for constructing a smart orchard expert knowledge base based on machine learning according to claim 3, wherein: The methods for obtaining orchard management decisions are: For the user's decision-making problems, based on the historical data in the expert knowledge base, a decision tree algorithm is used to classify the knowledge to obtain knowledge sets of different categories; Based on different categories of knowledge sets, the support vector machine algorithm is used to establish a knowledge prediction model to obtain prediction results; The prediction results are optimized based on a genetic algorithm to obtain intelligent decision-making suggestions for orchard management.

6. A system for constructing a smart orchard expert knowledge base based on machine learning, for implementing the method according to any one of claims 1 to 5, characterized in that: include: Data acquisition module, used to collect and pre-process orchard-related data to obtain an initial knowledge base; A knowledge standardization module, configured to construct a knowledge representation model based on the initial knowledge base using an ontology construction method, and obtain standardized knowledge representation based on the knowledge representation model; A knowledge reasoning chain acquisition module, configured to obtain a knowledge reasoning chain based on the standardized knowledge representation and using machine learning reasoning; An incremental module is used to learn new planting cases and expert feedback based on the knowledge reasoning chain using an incremental learning algorithm, automatically update the initial knowledge base, and obtain an expert knowledge base; The intelligent decision-making module is used to obtain orchard management decisions based on the expert knowledge base and adopt a machine learning algorithm to complete the construction of a smart orchard expert knowledge base based on machine learning.

Images