Water, fertilizer and irrigation management method and system based on agricultural big data and ai fusion

Abstract

The application provides a water and fertilizer irrigation management method and system based on agricultural big data and AI fusion, relates to the technical field of knowledge graph, constructs a knowledge graph of water and fertilizer irrigation, extracts semantic embedding vectors of soil moisture content changes between adjacent irrigation events, incrementally updates the knowledge graph for moisture content state transition, obtains an updated knowledge graph, analyzes water and fertilizer stress of field crops, obtains water and fertilizer stress levels of field crops in the current planting season, further updates the push weight of a water and fertilizer irrigation information subgraph associated with the current planting season in the updated knowledge graph, and determines the recommended confidence of each irrigation scheme in the current planting season; when a water and fertilizer irrigation suggestion request is initiated by a farmer terminal, an irrigation scheme is selected and pushed to the farmer terminal. The application can realize dynamic updating of the water and fertilizer irrigation management knowledge graph, complete accurate information pushing for different planting scenes, and thus improve the field operation adaptation practicability of the knowledge graph.

Description

Technical Field

[0001] This application relates to the field of knowledge graph technology, and more specifically, to a water and fertilizer irrigation management method and system based on the integration of agricultural big data and AI. Background Technology

[0002] With the continuous advancement of digital transformation in agriculture, knowledge graph technology, as a core technology for integrating multi-source agricultural data and realizing structured knowledge expression, is increasingly demonstrating its application value in the field of water and fertilizer irrigation management. It can break down the barriers of scattered and disorganized agricultural data, link and integrate relevant information such as soil, crops, and irrigation to form a systematic knowledge system, providing scientific support for water and fertilizer irrigation decisions, helping to solve problems such as resource waste and blind decision-making in traditional irrigation, and promoting the development of agriculture towards precision and efficiency. It is an important technical means to improve the level of agricultural production management.

[0003] In existing technologies, traditional water and fertilizer irrigation management models rely heavily on fragmented and scattered knowledge. Even with knowledge graph technology, it's difficult to achieve precise adaptation to actual field planting scenarios. Most existing knowledge graphs are statically constructed and cannot be updated to reflect dynamic changes in soil moisture and crop growth. Furthermore, they lack effective information delivery and adaptation mechanisms, resulting in weak targeting of irrigation-related knowledge delivery. This fails to meet the differentiated needs of different planting scenarios and growth stages, hindering the full realization of the practical guidance value of knowledge graphs and limiting their practical application in field irrigation management. Therefore, how to achieve dynamic updates to water and fertilizer irrigation management knowledge graphs and accurately deliver information to different planting scenarios to improve their practical applicability in the field has become a challenge for the industry. Summary of the Invention

[0004] This application provides a water and fertilizer irrigation management method and system based on the integration of agricultural big data and AI, which can realize dynamic updates of the water and fertilizer irrigation management knowledge graph and complete accurate information push for different planting scenarios, thereby improving the practical applicability of the knowledge graph in the field.

[0005] Firstly, this application provides a water and fertilizer irrigation management method based on the integration of agricultural big data and AI, the water and fertilizer irrigation management method comprising the following steps: Acquire agricultural big data and construct a knowledge graph of water and fertilizer irrigation; Semantic embedding vectors of soil moisture changes between adjacent irrigation events are extracted from water and fertilizer irrigation events under multiple consecutive planting seasons. The knowledge graph is then incrementally updated based on the soil moisture state transitions according to all the semantic embedding vectors to obtain the updated knowledge graph. Real-time soil data of field crops under the current planting season is collected and uploaded to the cloud. The water and fertilizer stress of field crops is analyzed by the artificial intelligence model in the cloud and the soil data to obtain the water and fertilizer stress level of field crops under the current planting season. The water and fertilizer stress levels are used to push and update the weight of the water and fertilizer irrigation information subgraph associated with the current planting season in the updated knowledge graph, thereby determining the recommendation confidence of each irrigation scheme under the current planting season. When farmers initiate a request for water and fertilizer irrigation suggestions through their terminals, an irrigation plan is selected based on the confidence level of all recommendations and pushed to the farmer's terminal.

[0006] In this embodiment, acquiring agricultural big data and constructing a knowledge graph for water and fertilizer irrigation specifically includes: Acquire agricultural big data, which includes crop variety data, soil type data, historical meteorological data, historical irrigation data, and historical fertilization data; Entity extraction and relation extraction are performed on the agricultural big data to obtain multiple entities related to water and fertilizer irrigation and the relationships between these entities. The entities include crops, soil types, irrigation methods, fertilizer types, irrigation time nodes, and corresponding water and fertilizer ratios. Construct the initial structure of the knowledge graph for water and fertilizer irrigation based on all entities and the relationships between them; The irrigation and fertilization history records from the agricultural big data are used as attribute information to fill the corresponding entity nodes in the initial structure of the knowledge graph, thus obtaining the knowledge graph of water and fertilizer irrigation.

[0007] In this embodiment, extracting the semantic embedding vector of soil moisture changes between adjacent irrigation events from water and fertilizer irrigation events across multiple consecutive planting seasons specifically includes: Water and fertilizer irrigation events for multiple consecutive planting seasons are obtained from historical databases. Each water and fertilizer irrigation event records the irrigation time, irrigation amount, fertilizer amount, and soil moisture parameters before and after irrigation. The soil moisture parameters after the end of the first irrigation event in two adjacent irrigation events are taken as the initial moisture state, and the soil moisture parameters before the start of the second irrigation event are taken as the ending moisture state. Calculate the change in moisture parameters between the initial moisture state and the final moisture state to obtain the moisture change vector between two adjacent water and fertilizer irrigation events; The water and fertilizer change vector between two adjacent water and fertilizer irrigation events is determined based on the irrigation time, irrigation amount, and fertilizer application amount of the previous and subsequent water and fertilizer irrigation events. The initial soil moisture state, the final soil moisture state, the soil moisture change vector, and the water and fertilizer change vector are semantically encoded to generate a semantic embedding vector of soil moisture change between adjacent irrigation events.

[0008] In this embodiment, the knowledge graph is incrementally updated based on all semantic embedding vectors to achieve moisture state transitions, resulting in the updated knowledge graph specifically including: Temporal parsing is performed on each semantic embedding vector to extract the soil moisture baseline state before irrigation, the soil moisture recovery state after irrigation, and the soil moisture evolution trajectory between the two. Based on the baseline moisture state and the moisture recovery state, node anchoring is performed in the knowledge graph to locate the corresponding moisture state transition path node pairs; The soil moisture evolution trajectory is encoded as a transition edge attribute and mapped to the nodes of the soil moisture state transition path to form an incremental edge for soil moisture state transition; Traverse all semantic embedding vectors and integrate the incremental edges parsed from each semantic embedding vector into the knowledge graph in the order of planting seasons to construct a cross-seasonal soil moisture state transition chain. The connectivity of the transition paths of the fused knowledge graph is verified based on the moisture state transition chain. Multiple incremental edges with the same start node and end node are semantically aggregated to obtain the updated knowledge graph.

[0009] In this embodiment, the connectivity of the transition paths of the fused knowledge graph is verified according to the moisture state transition chain. Multiple incremental edges with the same start node and end node are semantically aggregated to obtain the updated knowledge graph, which specifically includes: Using each transition node in the soil moisture state transition chain as the starting point of the traversal, the traversal is performed along the directed edges in the fused knowledge graph to mark all connected paths that start from the starting node and can reach the ending node. Filter out path groups with the same start node and the same end node from all connected paths, and extract the soil moisture evolution trajectory code carried by each incremental edge in each path group. All soil moisture evolution trajectory codes in each path are concatenated in chronological order to form a composite trajectory code. This composite trajectory code is then used to replace multiple incremental edges in that path, forming a single aggregated edge, thus obtaining the updated knowledge graph.

[0010] In this embodiment, water and fertilizer stress analysis of field crops is performed using a cloud-based artificial intelligence model and the soil data to obtain the specific water and fertilizer stress levels of field crops in the current planting season, including: The artificial intelligence model that has undergone transfer learning in advance is retrieved from the cloud, wherein the artificial intelligence model is a water and fertilizer stress recognition model; Determine the crop water tolerance threshold curve and crop nutrient tolerance threshold curve that match the crop variety and current planting stage in the field; The soil data is input into the water and fertilizer stress identification model. The water and nutrient deviation of the current soil moisture is determined by the identification results of the water and fertilizer stress identification model, the crop water tolerance threshold curve, and the crop nutrient tolerance threshold curve. The stress level is mapped based on the moisture deviation and the nutrient deviation, and the stress interval that the moisture deviation and the nutrient deviation fall into is determined as the water and fertilizer stress level of the field crop in the current planting season.

[0011] In this embodiment, the soil data is input into the water and fertilizer stress identification model. The determination of the current soil moisture deviation and nutrient deviation based on the identification results of the water and fertilizer stress identification model, the crop water tolerance threshold curve, and the crop nutrient tolerance threshold curve specifically includes: The soil data is converted into a soil feature vector, and the soil feature vector is input into the water and fertilizer stress identification model; The water and fertilizer stress confidence level output by the water and fertilizer stress identification model and the crop water tolerance threshold curve are used to perform water deviation analysis on the current soil moisture to obtain the water deviation degree of the current soil moisture. Nutrient deviation analysis of the current soil moisture is performed using the water and fertilizer stress confidence score output by the water and fertilizer stress identification model and the crop nutrient tolerance threshold curve to obtain the nutrient deviation degree of the current soil moisture.

[0012] In this embodiment, the water and fertilizer stress level is used to push and update the weight of the water and fertilizer irrigation information subgraph associated with the current planting season in the updated knowledge graph, thereby determining the recommendation confidence of each irrigation scheme under the current planting season. Specifically, this includes: Using the crop type identifier and growth stage identifier of the current planting season as the retrieval index, information subgraphs are extracted from the updated knowledge graph to obtain water and fertilizer irrigation information subgraphs associated with the current planting season. The weighted incentive coefficients are determined based on the stress interval in which the water and fertilizer stress levels are located; Traverse each irrigation scheme entity node in the water and fertilizer irrigation information subgraph, evaluate the path matching degree between the expected soil moisture response path of each irrigation scheme entity node and the current soil moisture status collected in real time, and obtain the path matching degree. The original push weights of the irrigation scheme entity nodes are adjusted in combination based on the path matching degree and the weight incentive coefficient to obtain the updated push weights of the irrigation scheme entity nodes. A recommendation confidence distribution is constructed using the updated push weights of all irrigation scheme entity nodes, and then the recommendation confidence of each irrigation scheme under the current planting season is determined based on the recommendation confidence distribution.

[0013] In this embodiment, each irrigation scheme entity node in the water and fertilizer irrigation information subgraph is traversed, and the expected soil moisture response path of each irrigation scheme entity node is evaluated for path consistency with the real-time collected current soil moisture status. The path consistency specifically includes: Take out an irrigation scheme entity node from the water and fertilizer irrigation information subgraph in sequence, and read the expected soil moisture response path stored in the irrigation scheme entity node. The expected soil moisture response path contains multiple expected soil moisture status nodes arranged in chronological order. Extract the current soil moisture value sequence from the real-time collected current soil moisture status, and align the current soil moisture value sequence with multiple expected soil moisture status nodes in the expected soil moisture response path according to time. Each moisture value in the current moisture value sequence after alignment is compared with the moisture value in the expected moisture state node at the corresponding time position. The number of matching pairs with differences within the preset tolerance range in all comparison results is counted. The ratio of the number of matching pairs to the total number of comparisons is used as the path fit.

[0014] Secondly, this application provides a water and fertilizer irrigation management system based on the integration of agricultural big data and AI, used to execute a water and fertilizer irrigation management method based on the integration of agricultural big data and AI, the water and fertilizer irrigation management system comprising: The knowledge graph construction module is used to acquire agricultural big data and construct a knowledge graph for water and fertilizer irrigation. The soil moisture semantic vector extraction and knowledge graph incremental update module is used to extract the semantic embedding vector of soil moisture changes between adjacent irrigation events from water and fertilizer irrigation events under multiple consecutive planting seasons, and to perform incremental updates of the knowledge graph based on all the semantic embedding vectors to obtain the updated knowledge graph. The crop water and fertilizer stress level analysis module is used to collect soil data of field crops in real time during the current planting season and upload it to the cloud. The cloud-based artificial intelligence model and the soil data are used to analyze the water and fertilizer stress of field crops and obtain the water and fertilizer stress level of field crops in the current planting season. The irrigation scheme confidence dynamic calibration module is used to push the weight update of the water and fertilizer irrigation information subgraph associated with the current planting season in the updated knowledge graph based on the water and fertilizer stress level, thereby determining the recommended confidence of each irrigation scheme under the current planting season. The precise push module for water and fertilizer irrigation plans is used to select an irrigation plan based on all recommendation confidence levels and push it to the farmer's terminal when the farmer initiates a water and fertilizer irrigation suggestion request through the terminal.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The system acquires agricultural big data and constructs a knowledge graph of water and fertilizer irrigation. Semantic embedding vectors representing soil moisture changes between adjacent irrigation events are extracted from water and fertilizer irrigation events across multiple consecutive planting seasons. The knowledge graph is then incrementally updated based on these semantic embedding vectors to reflect soil moisture state transitions, resulting in an updated knowledge graph. Real-time soil data of field crops during the current planting season is collected and uploaded to the cloud. A cloud-based artificial intelligence model and the soil data are used to analyze water and fertilizer stress in the field crops, determining the water and fertilizer stress level for the crops during the current planting season. The water and fertilizer stress level is used to update the push weights of the water and fertilizer irrigation information subgraph associated with the current planting season in the updated knowledge graph, thereby determining the recommendation confidence level for each irrigation scheme during the current planting season. When a farmer initiates a water and fertilizer irrigation suggestion request through their terminal, an irrigation scheme is selected based on all recommendation confidence levels and pushed to the farmer's terminal.

[0016] Therefore, this application demonstrates that irrigation schemes can be selected and pushed to farmers' terminals based on all recommendation confidence levels. Firstly, by acquiring agricultural big data and constructing a knowledge graph of water and fertilizer irrigation, the fragmented and scattered nature of related knowledge in traditional water and fertilizer irrigation management can be broken down. This systematically integrates multi-source information such as soil, crops, and irrigation, forming a complete knowledge system, effectively solving the problem of insufficient knowledge integration and difficulty in supporting scientific decision-making in existing technologies. Secondly, semantic embedding vectors of soil moisture changes are extracted from water and fertilizer irrigation events across multiple planting seasons, and incremental updates of the knowledge graph based on moisture state transitions are performed. This breaks the limitations of static knowledge graph construction, enabling the knowledge graph to iteratively optimize in real time according to the dynamic changes in field soil moisture, ensuring that the knowledge graph always aligns with actual field planting conditions, thus solving the technical shortcoming of static graphs being unable to adapt to dynamic changes in the field. Furthermore, by collecting field soil data for the current planting season in real time and uploading it to the cloud, cloud-based artificial intelligence can be utilized... The model can perform water and fertilizer stress analysis to obtain stress levels, accurately quantifying the current water and fertilizer requirements of crops. This provides a scientific and quantitative basis for the precise delivery of subsequent irrigation information, avoiding the problem of blind delivery. Next, by updating the delivery weights of the corresponding irrigation information subgraphs in the knowledge graph based on the water and fertilizer stress levels and determining the recommendation confidence of each irrigation scheme, an effective information delivery adaptation mechanism is established. This mechanism can achieve hierarchical adaptation of irrigation information according to the differentiated needs of different planting scenarios and crop growth stages, solving the deficiency of insufficient targeting of delivery content in existing technologies. Finally, when farmers initiate irrigation suggestion requests, the appropriate irrigation scheme is selected based on the recommendation confidence and delivered to the farmer's terminal. This transforms the theoretical knowledge of the knowledge graph into practical guidance for front-line field operations, fully leveraging the practical guidance value of the knowledge graph. This effectively reduces resource waste and decision-making blindness in traditional irrigation, improves the practical application effect of the knowledge graph in field irrigation management, and promotes the development of agricultural water and fertilizer irrigation towards precision and efficiency.

[0017] In summary, the technical solution adopted in this application can realize the dynamic updating of the knowledge graph of water and fertilizer irrigation management, and complete the accurate information push for different planting scenarios, thereby improving the practical applicability of the knowledge graph in the field. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1This is an exemplary flowchart of a water and fertilizer irrigation management method based on the integration of agricultural big data and AI, provided in this application. Figure 2 This is an application scenario diagram of the water and fertilizer irrigation management system provided in this application; Figure 3 This is a flowchart illustrating the process for determining the level of water and fertilizer stress provided in this application; Figure 4 This is a modular structure diagram of a water and fertilizer irrigation management system based on the integration of agricultural big data and AI, provided in this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] This application provides a water and fertilizer irrigation management method and system based on the integration of agricultural big data and AI. Its core is to acquire agricultural big data and construct a knowledge graph for water and fertilizer irrigation; extract semantic embedding vectors of soil moisture changes between adjacent irrigation events from water and fertilizer irrigation events across multiple consecutive planting seasons; incrementally update the knowledge graph based on the moisture state transitions using all semantic embedding vectors to obtain an updated knowledge graph; collect soil data of field crops in the current planting season in real time and upload it to the cloud; perform water and fertilizer stress analysis on field crops using cloud-based AI models and the soil data to obtain the water and fertilizer stress level of field crops in the current planting season; update the push weight of the water and fertilizer irrigation information subgraph associated with the current planting season in the updated knowledge graph based on the water and fertilizer stress level, thereby determining the recommendation confidence of each irrigation scheme in the current planting season; when a farmer initiates a water and fertilizer irrigation suggestion request through a terminal, an irrigation scheme is selected based on all recommendation confidence levels and pushed to the farmer's terminal.

[0022] Example 1: To better understand the above technical solution, the following will provide a detailed description of the technical solution in conjunction with the accompanying drawings and specific implementation methods. (Refer to...) Figure 1 As shown in the figure, this is an exemplary flowchart of a water and fertilizer irrigation management method based on the integration of agricultural big data and AI according to this embodiment of the application. The water and fertilizer irrigation management method includes the following steps: In step S1, agricultural big data is acquired and a knowledge graph of water and fertilizer irrigation is constructed.

[0023] In this embodiment, acquiring agricultural big data and constructing a knowledge graph for water and fertilizer irrigation can be achieved through the following steps: Acquire agricultural big data, which includes crop variety data, soil type data, historical meteorological data, historical irrigation data, and historical fertilization data; Entity extraction and relation extraction are performed on the agricultural big data to obtain multiple entities related to water and fertilizer irrigation and the relationships between these entities. The entities include crops, soil types, irrigation methods, fertilizer types, irrigation time nodes, and corresponding water and fertilizer ratios. Construct the initial structure of the knowledge graph for water and fertilizer irrigation based on all entities and the relationships between them; The irrigation and fertilization history records from the agricultural big data are used as attribute information to fill the corresponding entity nodes in the initial structure of the knowledge graph, thus obtaining the knowledge graph of water and fertilizer irrigation.

[0024] It should be noted that the agricultural big data mentioned in this application refers to basic data related to irrigation and fertilization in all dimensions of field planting; the entities related to water and fertilizer irrigation are data node units with independent semantic information within a knowledge graph; the relationships between the entities are semantic connection links between the data nodes; and the knowledge graph of water and fertilizer irrigation is a structured data network used for soil moisture updates and water and fertilizer irrigation recommendations.

[0025] In specific implementation, firstly, stored data is synchronously retrieved from multiple data sources, including agricultural cloud databases, field IoT data collection terminals, and public meteorological databases. Crop variety data, soil type data, historical meteorological data, irrigation history data, and fertilization history data are extracted sequentially, and the integrated multi-source data is used as agricultural big data. Secondly, the BERT entity joint extraction algorithm is employed to complete text data traversal, word segmentation, and entity boundary recognition based on a pre-defined agricultural entity dictionary. This sequentially identifies all entity nodes related to crops, soil types, irrigation methods, fertilizer types, irrigation time nodes, and water-fertilizer ratios. Simultaneously, semantic dependency analysis rules are used to identify the adaptation correspondence, application correspondence, and temporal correspondence between entities. All identified nodes and link information are then processed... This involves identifying multiple entities related to water and fertilizer irrigation and their interrelationships. Next, using all extracted entities as independent node units and the interrelationships as undirected edges between nodes, the spatial topology of all nodes and edges is arranged according to graph topology construction rules. This completed topological framework serves as the initial structure of the water and fertilizer irrigation knowledge graph. Finally, all entity nodes within the initial knowledge graph structure are traversed, and the corresponding data category labels for each node are matched. Irrigation and fertilization history data stored in the agricultural big data are then aligned and mounted one by one into the data attribute fields of the corresponding category entity nodes, completing all attribute data filling operations. The complete topological data structure after attribute filling is then used as the water and fertilizer irrigation knowledge graph.

[0026] It should also be noted that in this application, references Figure 2 As shown in the figure, this is an application scenario diagram of the water and fertilizer irrigation management system provided in this application embodiment. The water and fertilizer irrigation management system uses a network as its core hub, connecting drone terminals and satellite equipment at the upper layer to acquire remote sensing data and meteorological information from a large area of ​​farmland, realizing the collection of multi-source agricultural data. The ground control terminal is used for unified scheduling and data aggregation of field IoT devices, collecting real-time soil moisture data including soil moisture, conductivity, and nutrient content. The airspace management platform on the right corresponds to the cloud processing center, internally integrating airspace management servers, data storage servers, and communication servers, used to complete the storage of agricultural big data, knowledge graph construction, and water and fertilizer stress analysis based on artificial intelligence models, and to dynamically update and infer the knowledge graph. The user terminals below include mobile phones, tablets, laptops, and desktop computers, used by farmers to initiate irrigation decision requests and receive irrigation plans pushed by the system. The whole system forms a closed loop of data collection—cloud analysis—knowledge graph update—plan recommendation—terminal feedback, realizing precise water and fertilizer irrigation decisions and intelligent information push for different planting scenarios.

[0027] In step S2, semantic embedding vectors of soil moisture changes between adjacent irrigation events are extracted from water and fertilizer irrigation events under multiple consecutive planting seasons. The knowledge graph is incrementally updated based on the soil moisture state transition according to all the semantic embedding vectors to obtain the updated knowledge graph.

[0028] In this embodiment, the semantic embedding vector of soil moisture changes between adjacent irrigation events can be extracted from water and fertilizer irrigation events over multiple consecutive planting seasons using the following steps: Water and fertilizer irrigation events for multiple consecutive planting seasons are obtained from historical databases. Each water and fertilizer irrigation event records the irrigation time, irrigation amount, fertilizer amount, and soil moisture parameters before and after irrigation. The soil moisture parameters after the end of the first irrigation event in two adjacent irrigation events are taken as the initial moisture state, and the soil moisture parameters before the start of the second irrigation event are taken as the ending moisture state. Calculate the change in moisture parameters between the initial moisture state and the final moisture state to obtain the moisture change vector between two adjacent water and fertilizer irrigation events; The water and fertilizer change vector between two adjacent water and fertilizer irrigation events is determined based on the irrigation time, irrigation amount, and fertilizer application amount of the previous and subsequent water and fertilizer irrigation events. The initial soil moisture state, the final soil moisture state, the soil moisture change vector, and the water and fertilizer change vector are semantically encoded to generate a semantic embedding vector of soil moisture change between adjacent irrigation events.

[0029] It should be noted that, in this application, the water and fertilizer irrigation event refers to information recording water and fertilizer irrigation-related parameters; the initial soil moisture state refers to the initial soil moisture information between two adjacent water and fertilizer irrigation intervals; the final soil moisture state refers to the final soil moisture information between two adjacent water and fertilizer irrigation intervals; the soil moisture change vector is a vector that quantifies the difference between the initial and final soil moisture; and the semantic embedding vector of soil moisture change between adjacent irrigation events represents a semantic encoding vector that measures the soil moisture change between adjacent water and fertilizer irrigation events.

[0030] In practice, the process begins by retrieving all irrigation and fertilization records from a historical database storing agricultural big data, covering multiple consecutive planting seasons. For each record, the irrigation time, irrigation volume, fertilizer application amount, and soil moisture parameters before and after irrigation are extracted. These soil moisture parameters specifically include soil moisture, soil electrical conductivity, and nitrogen, phosphorus, and potassium ion concentrations in the root zone. Each record containing all these parameters is considered a single irrigation and fertilization event. Next, all irrigation and fertilization events are sorted chronologically. Adjacent irrigation events are then extracted sequentially. The soil moisture parameters at the end of the previous irrigation event are used as the initial moisture state, and the soil moisture parameters at the beginning of the next irrigation event are used as the ending moisture state. These two sets of parameters are then used as the initial and ending moisture states, respectively. Finally, the soil moisture and soil electrical conductivity parameters in the initial and ending moisture states are analyzed. The differences in nitrogen, phosphorus, and potassium ion concentrations in the root zone were calculated separately. These three differences were then integrated into a multidimensional data set in the order of soil moisture, soil electrical conductivity, and root zone nitrogen, phosphorus, and potassium ion concentrations. This multidimensional data set was used as the soil moisture change vector. Next, the differences in irrigation time, irrigation amount, and fertilizer application amount between two adjacent irrigation events were calculated, and the vector composed of these differences was used as the water and fertilizer change vector between the two adjacent irrigation events. Finally, the Word2Vec semantic coding algorithm was used. The soil moisture change vector, water and fertilizer change vector, initial soil moisture state, and final soil moisture state between two adjacent irrigation events were used as algorithm inputs. The algorithm performed semantic mapping and low-dimensional compression on the input data, and the low-dimensional dense feature vector output by the algorithm was used as the semantic embedding vector of soil moisture change between adjacent irrigation events.

[0031] In this embodiment, the knowledge graph is incrementally updated based on all semantic embedding vectors to achieve the updated knowledge graph. This can be achieved through the following steps: Temporal parsing is performed on each semantic embedding vector to extract the soil moisture baseline state before irrigation, the soil moisture recovery state after irrigation, and the soil moisture evolution trajectory between the two. Based on the baseline moisture state and the moisture recovery state, node anchoring is performed in the knowledge graph to locate the corresponding moisture state transition path node pairs; The soil moisture evolution trajectory is encoded as a transition edge attribute and mapped to the nodes of the soil moisture state transition path to form an incremental edge for soil moisture state transition; Traverse all semantic embedding vectors and integrate the incremental edges parsed from each semantic embedding vector into the knowledge graph in the order of planting seasons to construct a cross-seasonal soil moisture state transition chain. The connectivity of the transition paths of the fused knowledge graph is verified based on the moisture state transition chain. Multiple incremental edges with the same start node and end node are semantically aggregated to obtain the updated knowledge graph.

[0032] It should be noted that, in this application, the soil moisture baseline state refers to the initial soil moisture information before irrigation; the soil moisture recovery state refers to the final soil moisture information after irrigation; the soil moisture evolution trajectory encoding is encoded information representing the process of soil moisture change from the baseline state to the recovery state; the soil moisture state transition path node pair is a node pair in the knowledge graph consisting of two nodes corresponding to the soil moisture baseline state and the soil moisture recovery state; the incremental edge is a connection edge carrying the soil moisture evolution trajectory encoding to supplement the state transition relationship of the knowledge graph; the cross-seasonal soil moisture state transition chain is a coherent state transition link formed by merging all incremental edges in the order of planting season; and the updated knowledge graph is a structured data network after connectivity verification and semantic aggregation, and after supplementing the complete soil moisture state transition relationship.

[0033] In specific implementation, firstly, the LSTM temporal parsing algorithm is used to decompose each semantic embedding vector temporally, extracting the baseline soil moisture state before irrigation and the recovery soil moisture state after irrigation vector by vector. Simultaneously, the temporal features of soil moisture changes between the two are extracted and encoded, with the encoding result serving as the soil moisture evolution trajectory encoding. Secondly, the cosine similarity matching algorithm is used to calculate the similarity between the extracted baseline and recovery soil moisture states and the attribute information of all soil moisture-related nodes in the knowledge graph. Nodes with similarity reaching a preset threshold are designated as baseline and recovery nodes, respectively, and the combination of these two nodes is taken as the corresponding soil moisture state transition path node pair. Next, the empty spaces between the soil moisture state transition path node pairs in the knowledge graph are read. For white connecting edges, the extracted soil moisture evolution trajectory is encoded and written into the attribute field of the blank connecting edge to complete the attribute mapping operation. The connecting edge carrying the soil moisture evolution trajectory encoding is used as the incremental edge for soil moisture state transition. Then, all semantic embedding vectors are traversed in chronological order of planting season. The incremental edges parsed from each semantic embedding vector are added sequentially to the corresponding node pairs in the knowledge graph to ensure that the order of adding incremental edges is consistent with the planting season. The coherent link formed by connecting all incremental edges is used as the cross-seasonal soil moisture state transition chain. Finally, the connectivity of the transition path is checked on the fused knowledge graph according to the soil moisture state transition chain. Multiple incremental edges with the same start node and end node are semantically aggregated to obtain the updated knowledge graph.

[0034] In this embodiment, the connectivity of the transition paths of the fused knowledge graph is verified according to the soil moisture state transition chain. Multiple incremental edges with the same start node and end node are semantically aggregated to obtain the updated knowledge graph. This can be achieved through the following steps: Using each transition node in the soil moisture state transition chain as the starting point of the traversal, the traversal is performed along the directed edges in the fused knowledge graph to mark all connected paths that start from the starting node and can reach the ending node. Filter out path groups with the same start node and the same end node from all connected paths, and extract the soil moisture evolution trajectory code carried by each incremental edge in each path group. All soil moisture evolution trajectory codes in each path are concatenated in chronological order to form a composite trajectory code. This composite trajectory code is then used to replace multiple incremental edges in that path, forming a single aggregated edge, thus obtaining the updated knowledge graph.

[0035] It should be noted that the soil moisture state transition chain described in this application is a cross-seasonal soil moisture temporal link formed after merging all incremental edges; the connected path is a complete link that can be fully traversed from a graph node along a directed edge to the target node; the path group is a set of multiple connected paths with completely identical starting and ending nodes; the composite trajectory encoding is a comprehensive feature encoding after sequentially splicing and integrating multiple sets of temporal encodings; the aggregated edge is a single-attribute connecting edge formed after replacing multiple incremental edges; and the updated knowledge graph is a water and fertilizer irrigation knowledge graph after completing connectivity verification and semantic aggregation optimization.

[0036] In specific implementation, firstly, a depth-first search algorithm is used, taking all transition nodes in the soil moisture state transition chain as the starting node for traversal, and performing a global traversal along all directed edges within the fused knowledge graph to fully record each complete link that can be successfully traversed. The complete traversable links recorded during the traversal are taken as the connected paths that start from the starting node and can reach the ending node. Secondly, all connected paths are classified and divided according to node matching rules, and the sets of paths with the same starting and ending nodes are selected to form corresponding path groups, thereby obtaining the soil moisture evolution trajectory code carried by each incremental edge in each group of paths. Finally, all soil moisture evolution trajectory codes in each group are sequentially connected and integrated according to the chronological order of the planting season. The integrated overall code data is taken as the composite trajectory code. The original multiple incremental edges in each group are deleted, and the composite trajectory code is attached to the corresponding nodes to form a single aggregate edge. The complete graph structure after all path groups have been replaced and optimized is taken as the updated knowledge graph.

[0037] In step S3, soil data of field crops under the current planting season is collected in real time and uploaded to the cloud. The water and fertilizer stress of field crops is analyzed by the artificial intelligence model in the cloud and the soil data to obtain the water and fertilizer stress level of field crops under the current planting season.

[0038] It should be noted that the soil data mentioned in this application refers to multi-dimensional soil parameters that can reflect the water and fertilizer status of crop growth in field plots; the cloud refers to an agricultural cloud service platform.

[0039] In practice, firstly, multi-parameter IoT soil sensors deployed in the field crop planting area continuously collect on-site data according to the preset collection frequency, simultaneously collecting parameters such as soil moisture, soil conductivity, soil temperature, and nitrogen, phosphorus, and potassium ion concentrations in the root zone. All collected parameter data are integrated and packaged as soil data for the field crops in the current planting season. Secondly, the collected soil data is uploaded to the cloud.

[0040] Preferably, in this embodiment, water and fertilizer stress analysis of field crops is performed using a cloud-based artificial intelligence model and the soil data to obtain the water and fertilizer stress level of the field crops in the current planting season, with reference to... Figure 3 As shown in the figure, this is a flowchart illustrating the process of determining the water and fertilizer stress level in some embodiments of this application. In this embodiment, the determination of the water and fertilizer stress level can be achieved by the following steps: In step S31, an artificial intelligence model that has completed transfer learning in advance is retrieved from the cloud, wherein the artificial intelligence model is a water and fertilizer stress recognition model; In step S32, the crop water tolerance threshold curve and crop nutrient tolerance threshold curve matching the crop variety and the current planting stage in the field are determined. In step S33, the soil data is input into the water and fertilizer stress identification model, and the current soil moisture deviation and nutrient deviation are determined by the identification results of the water and fertilizer stress identification model, the crop water tolerance threshold curve and the crop nutrient tolerance threshold curve. In step S34, stress level mapping is performed based on the moisture deviation and the nutrient deviation, and the stress interval into which the moisture deviation and nutrient deviation fall is determined as the water and fertilizer stress level of the field crop in the current planting season.

[0041] It should be noted that the water and fertilizer stress identification model described in this application is an artificial intelligence model pre-trained by transfer learning to determine the water and fertilizer stress state of crops. Specifically, it can be implemented as follows: Using a transfer learning algorithm, a publicly available large-scale multi-crop, multi-planting-stage water and fertilizer stress sample dataset (covering four categories of samples: normal, mild stress, moderate stress, and severe stress) is used as the transfer source data to train a basic water and fertilizer stress identification model. Then, measured water and fertilizer data and stress state labels of the current crop variety and each planting stage in the local field are selected as fine-tuning samples to adjust the model parameters. After fine-tuning the model, the model's discrimination accuracy is verified to be above 95%, confirming model convergence and stability. The model trained through transfer learning is then used as the water and fertilizer stress identification model. The crop water tolerance threshold curve represents a reference curve defining the reasonable range of soil moisture under normal crop growth conditions; the crop nutrient tolerance threshold curve represents a reference curve defining the reasonable range of soil nutrients under normal crop growth conditions.

[0042] In practice, firstly, the pre-learned artificial intelligence model, namely the water and fertilizer stress identification model, is retrieved from the cloud. Secondly, crop category identification and field records of crop growth surveys are retrieved. Using a keyword matching retrieval algorithm, water and fertilizer tolerance sampling data corresponding to the current crop variety throughout its entire growth period are retrieved. This water and fertilizer tolerance sampling data consists of monitoring records obtained through continuous field trials at different growth stages, reflecting the crop's soil moisture carrying capacity limit and nutrient tolerance critical range. Referring to well-known growth period division standards in crop cultivation, and combining this with plant appearance and growth characteristics, the real-time planting stage is determined. Multiple sets of water and fertilizer tolerance sampling data corresponding to the real-time planting stage were selected and extracted. A piecewise linear fitting algorithm was used to fit the multiple sets of water and fertilizer tolerance sampling data. The two continuous smooth curves formed after fitting were used as the crop water tolerance threshold curve and the crop nutrient tolerance threshold curve, respectively. Then, the soil data was input into the water and fertilizer stress identification model. The water and fertilizer stress identification model was used to extract the water deviation and nutrient deviation between the current soil moisture and the crop water and fertilizer tolerance threshold curve. Finally, the obtained water deviation and nutrient deviation were substituted into the following formula to calculate the water and fertilizer stress score of the field crop in the current planting season, that is: ,in, Score for water and fertilizer stress; This is a correction factor for the growth stage, specifically the jointing stage. =1.2, Grouting period =1.1, Maturity stage =0.9 (determined based on field trial data) , These are the weighting coefficients. =0.6、 =0.4 (Verified by multiple control experiments, this weighting ratio can most accurately reflect the degree of influence of water and fertilizer on crops). For moisture deviation, The upper limit of water tolerance for the current growth stage of the crop (determined through field trials, such as the jointing stage of wheat). =20%) Nutrient deviation, The base correction value is set at 5 in this application based on the blank control group experiment. Furthermore, after calculating the water and fertilizer stress score, the levels are classified according to industry standards. <30 points indicates no coercion, 30 ≤ <60 points indicates mild stress, 60 ≤ A score of <80 indicates moderate stress. A score of ≥80 indicates severe stress, and the corresponding level will be used as the water and fertilizer stress level of the crop in the current planting season.

[0043] In this embodiment, the soil data is input into the water and fertilizer stress identification model, and the current soil moisture deviation and nutrient deviation are determined by the identification results of the water and fertilizer stress identification model, the crop water tolerance threshold curve, and the crop nutrient tolerance threshold curve. This can be achieved through the following steps: The soil data is converted into a soil feature vector, and the soil feature vector is input into the water and fertilizer stress identification model; The water and fertilizer stress confidence level output by the water and fertilizer stress identification model and the crop water tolerance threshold curve are used to perform water deviation analysis on the current soil moisture to obtain the water deviation degree of the current soil moisture. Nutrient deviation analysis of the current soil moisture is performed using the water and fertilizer stress confidence score output by the water and fertilizer stress identification model and the crop nutrient tolerance threshold curve to obtain the nutrient deviation degree of the current soil moisture.

[0044] It should be noted that the water and fertilizer stress confidence level mentioned in this application is a confidence value that the current soil water and fertilizer status is under stress; the water deviation degree reflects the degree of adaptation between soil water and crop requirements; and the nutrient deviation degree reflects the degree of adaptation between soil nutrients and crop requirements.

[0045] In practice, firstly, the soil data uploaded to the cloud in real time is standardized. The four standardized parameters are then concatenated and integrated in the order of soil moisture, soil electrical conductivity, root zone nitrogen ion concentration, and root zone phosphorus and potassium ion concentration to form a soil feature vector. This soil feature vector is then input into the feature extraction layer of the water and fertilizer stress identification model. Secondly, the water and fertilizer stress identification model outputs the water and fertilizer stress confidence score, while simultaneously acquiring the crop water tolerance threshold curve and the current soil moisture condition. The actual soil moisture value in the current condition is then compared item by item with the upper and lower limits of the water tolerance threshold curve, and the difference between the actual value and the midpoint of the interval is calculated. The output water and fertilizer stress confidence score is used as a weighting coefficient, and the difference is then weighted and corrected. The corrected quantitative deviation result is used as the current soil moisture deviation. Next, the water and fertilizer stress confidence score is output through the water and fertilizer stress identification model. At the same time, the crop water tolerance threshold curve and the current soil moisture are obtained. The actual nutrient values ​​in the current soil moisture are compared with the upper and lower limit intervals of the nutrient values ​​corresponding to the crop water tolerance threshold curve. The difference between the actual value and the midpoint of the interval is calculated. The output water and fertilizer stress confidence score is used as a weighting coefficient, and the difference is then weighted and corrected. The corrected quantitative deviation result is used as the current soil moisture nutrient deviation.

[0046] In step S4, the water and fertilizer stress level is used to push and update the weight of the water and fertilizer irrigation information subgraph associated with the current planting season in the updated knowledge graph, thereby determining the recommendation confidence of each irrigation scheme under the current planting season.

[0047] In this embodiment, the recommendation confidence of each irrigation scheme under the current planting season is determined by pushing the weight update of the water and fertilizer irrigation information subgraph associated with the current planting season in the updated knowledge graph based on the water and fertilizer stress level, which can be achieved through the following steps: Using the crop type identifier and growth stage identifier of the current planting season as the retrieval index, information subgraphs are extracted from the updated knowledge graph to obtain water and fertilizer irrigation information subgraphs associated with the current planting season. The weighted incentive coefficients are determined based on the stress interval in which the water and fertilizer stress levels are located; Traverse each irrigation scheme entity node in the water and fertilizer irrigation information subgraph, evaluate the path matching degree between the expected soil moisture response path of each irrigation scheme entity node and the current soil moisture status collected in real time, and obtain the path matching degree. The original push weights of the irrigation scheme entity nodes are adjusted in combination based on the path matching degree and the weight incentive coefficient to obtain the updated push weights of the irrigation scheme entity nodes. A recommendation confidence distribution is constructed using the updated push weights of all irrigation scheme entity nodes, and then the recommendation confidence of each irrigation scheme under the current planting season is determined based on the recommendation confidence distribution.

[0048] It should be noted that the water and fertilizer irrigation information subgraph mentioned in this application is a local topological subnetwork within the updated knowledge graph, which is bound to the corresponding crop type, growth cycle, and stores all irrigation scheme-related nodes and path information; the weight incentive coefficient is a global adjustment parameter set according to different stress level ranges to regulate the adjustment range of irrigation scheme weights; the path matching degree is a feature value that quantifies the degree of matching between the preset soil moisture change path of the irrigation scheme and the real-time soil moisture status in the field; the original push weight is the basic weight value preset for each irrigation scheme entity node during the knowledge graph initialization stage; the updated push weight is the node weight value that is finally retained after multiple parameter composite corrections; the recommendation confidence distribution is an ordered weight sequence formed by sequentially arranging the updated weights of all nodes; and the recommendation confidence is the final quantitative feature obtained by combining the overall weight distribution characteristics and used to characterize the adaptation priority of irrigation schemes.

[0049] In specific implementation, firstly, the crop type identifier and growth stage identifier corresponding to the current planting season are retrieved from cloud storage. These two identifiers are combined to form a dedicated composite retrieval index. A subgraph extraction and traversal algorithm from the knowledge graph domain is used to retrieve and match node and edge information along the updated global topology of the knowledge graph. All nodes, state transition edges, and associated attribute information that completely match the index information are selected. The extracted and integrated local topology is then used as a subgraph of water and fertilizer irrigation information associated with the current planting season. Secondly, stress intervals at various levels (no stress, mild stress, moderate stress, severe stress) are obtained. The numerical boundary range of each interval is defined. The quantified values ​​of the water and fertilizer stress levels obtained in this study are substituted into the interval range for matching to determine the specific stress interval to which it belongs. An interval linear mapping algorithm is used, with the severity of the stress interval as the mapping benchmark, to map the interval level to the corresponding adjustment parameter. The mapping rules are based on well-known stress levels in the field of field water and fertilizer management. The forced regulation logic is set such that the higher the stress level, the larger the weight incentive coefficient. The regulation coefficient obtained from the mapping calculation is used as the weight incentive coefficient for this weight update process. Next, each irrigation scheme entity node in the water and fertilizer irrigation information subgraph is traversed, and the expected soil moisture response path of each irrigation scheme entity node is evaluated with the real-time collected current soil moisture status to obtain the path consistency. Then, the original push weight of the irrigation scheme entity node is obtained, and the product between the path consistency and the weight incentive coefficient is multiplied by the original push weight. The result of the multiplication is used as the updated push weight of the irrigation scheme entity node. Finally, according to the topological sorting order of the irrigation scheme entity nodes, the updated push weights corresponding to all irrigation scheme entity nodes are arranged in a regular manner to form a continuous and ordered set of weight sequences as the recommendation confidence distribution. The recommendation confidence of each irrigation scheme under the current planting season is then calculated using the following formula: ,in This refers to the position index of the updated push weight in the recommendation confidence distribution. Indicates the first The recommendation confidence level of the irrigation scheme corresponding to the updated location weight is used to push the information. This is a correction factor for the growth stage. For the recommendation confidence distribution, the th The updated push weight at the location level For the recommendation confidence distribution, the th The topological weight coefficient at the location is calculated as follows: The topological weight coefficient decreases sequentially according to the topological order. , This represents the total number of entity nodes in the irrigation scheme.

[0050] In this embodiment, each irrigation scheme entity node in the water and fertilizer irrigation information subgraph is traversed, and the expected soil moisture response path of each irrigation scheme entity node is evaluated against the real-time collected current soil moisture status. The path consistency can be obtained by the following steps: Take out an irrigation scheme entity node from the water and fertilizer irrigation information subgraph in sequence, and read the expected soil moisture response path stored in the irrigation scheme entity node. The expected soil moisture response path contains multiple expected soil moisture status nodes arranged in chronological order. Extract the current soil moisture value sequence from the real-time collected current soil moisture status, and align the current soil moisture value sequence with multiple expected soil moisture status nodes in the expected soil moisture response path according to time. Each moisture value in the current moisture value sequence after alignment is compared with the moisture value in the expected moisture state node at the corresponding time position. The number of matching pairs with differences within the preset tolerance range in all comparison results is counted. The ratio of the number of matching pairs to the total number of comparisons is used as the path fit.

[0051] It should be noted that, in this application, the irrigation scheme entity node is a topological node within the water and fertilizer irrigation information subgraph that carries all attribute data of a single irrigation scheme and the corresponding soil moisture change time series data; the expected soil moisture response path is the entire process of theoretical soil moisture changes over time after the implementation of the irrigation scheme; the expected soil moisture status node represents a node that stores theoretical soil moisture values; the current soil moisture value sequence is the actual soil moisture value sequence arranged in chronological order; the preset tolerance range represents the range for determining whether the deviation of the soil moisture value is within a reasonable fluctuation range; and the path matching degree represents the degree of matching between the theoretical soil moisture change process of the irrigation scheme and the actual soil moisture change process in the field.

[0052] In practical implementation, firstly, following the topological traversal order of the knowledge graph, each irrigation scheme entity node in the water and fertilizer irrigation information subgraph is extracted sequentially. The built-in data reading interface of each irrigation scheme entity node is then called to extract the expected soil moisture response path bound to each node. This path contains multiple soil moisture nodes arranged in chronological order (one node per hour, covering the entire irrigation cycle). Each soil moisture node contains two core pieces of information: a specific soil moisture value and a timestamp. Simultaneously, field soil moisture sensors are activated to collect actual soil moisture data at the same time intervals as the expected soil moisture response path, using linear interpolation. The algorithm completes the missing soil moisture data during the collection process, forming a current moisture value sequence that is completely consistent with the time interval of the expected moisture response path. Then, a dynamic time warping algorithm is used to perform a full time-series comparison between the expected moisture response path and the current moisture value sequence. The algorithm parameters are set as follows: the comparison window size is 3 consecutive time nodes, the step size is 1, and the difference between the expected moisture value and the actual moisture value at each time node is calculated one by one. If the absolute value of the difference is within the preset tolerance range, it is determined to be a match. The number of all successfully matched time nodes is counted, and then divided by the total number of time nodes. The value obtained by division is used as the path matching degree.

[0053] In step S5, when a farmer initiates a water and fertilizer irrigation suggestion request through the terminal, an irrigation plan is selected based on all the recommendation confidence levels and pushed to the farmer's terminal.

[0054] In practice, the cloud platform monitors the network data signals sent by the farmer's terminal in real time. When it receives the water and fertilizer irrigation suggestion request initiated by the farmer through the terminal, it selects the irrigation plan corresponding to the highest recommendation confidence and transmits the complete execution parameters of the irrigation plan (irrigation duration, water and fertilizer ratio, operation period) to the farmer's terminal. The terminal automatically receives the data and clearly displays the plan details on the interface for the farmer to refer to and execute.

[0055] Therefore, this application demonstrates that irrigation schemes can be selected and pushed to farmers' terminals based on all recommendation confidence levels. Firstly, by acquiring agricultural big data and constructing a knowledge graph of water and fertilizer irrigation, the fragmented and scattered nature of related knowledge in traditional water and fertilizer irrigation management can be broken down. This systematically integrates multi-source information such as soil, crops, and irrigation, forming a complete knowledge system, effectively solving the problem of insufficient knowledge integration and difficulty in supporting scientific decision-making in existing technologies. Secondly, semantic embedding vectors of soil moisture changes are extracted from water and fertilizer irrigation events across multiple planting seasons, and incremental updates of the knowledge graph based on moisture state transitions are performed. This breaks the limitations of static knowledge graph construction, enabling the knowledge graph to iteratively optimize in real time according to the dynamic changes in field soil moisture, ensuring that the knowledge graph always aligns with actual field planting conditions, thus solving the technical shortcoming of static graphs being unable to adapt to dynamic changes in the field. Furthermore, by collecting field soil data for the current planting season in real time and uploading it to the cloud, cloud-based artificial intelligence can be utilized... The model can perform water and fertilizer stress analysis to obtain stress levels, accurately quantifying the current water and fertilizer requirements of crops. This provides a scientific and quantitative basis for the precise delivery of subsequent irrigation information, avoiding the problem of blind delivery. Next, by updating the delivery weights of the corresponding irrigation information subgraphs in the knowledge graph based on the water and fertilizer stress levels and determining the recommendation confidence of each irrigation scheme, an effective information delivery adaptation mechanism is established. This mechanism can achieve hierarchical adaptation of irrigation information according to the differentiated needs of different planting scenarios and crop growth stages, solving the deficiency of insufficient targeting of delivery content in existing technologies. Finally, when farmers initiate irrigation suggestion requests, the appropriate irrigation scheme is selected based on the recommendation confidence and delivered to the farmer's terminal. This transforms the theoretical knowledge of the knowledge graph into practical guidance for front-line field operations, fully leveraging the practical guidance value of the knowledge graph. This effectively reduces resource waste and decision-making blindness in traditional irrigation, improves the practical application effect of the knowledge graph in field irrigation management, and promotes the development of agricultural water and fertilizer irrigation towards precision and efficiency.

[0056] In summary, the technical solution adopted in this application can realize the dynamic updating of the knowledge graph of water and fertilizer irrigation management, and complete the accurate information push for different planting scenarios, thereby improving the practical applicability of the knowledge graph in the field.

[0057] Example 2: This application provides a water and fertilizer irrigation management system based on the integration of agricultural big data and AI, referencing... Figure 4 As shown in the figure, this is a modular structure diagram of a water and fertilizer irrigation management system based on the integration of agricultural big data and AI according to this embodiment of the present application. The water and fertilizer irrigation management system includes: Knowledge graph construction module 100 is used to acquire agricultural big data and construct a knowledge graph for water and fertilizer irrigation. The soil moisture semantic vector extraction and knowledge graph incremental update module 200 is used to extract the semantic embedding vector of soil moisture changes between adjacent irrigation events from water and fertilizer irrigation events under multiple consecutive planting seasons, and to perform incremental updates of the knowledge graph based on all the semantic embedding vectors to obtain the updated knowledge graph. The crop water and fertilizer stress level analysis module 300 is used to collect soil data of field crops in the current planting season in real time and upload it to the cloud. The water and fertilizer stress of field crops is analyzed by the artificial intelligence model in the cloud and the soil data to obtain the water and fertilizer stress level of field crops in the current planting season. The irrigation scheme confidence dynamic calibration module 400 is used to push the weight update of the water and fertilizer irrigation information subgraph associated with the current planting season in the updated knowledge graph based on the water and fertilizer stress level, thereby determining the recommended confidence of each irrigation scheme under the current planting season. The 500 precise water and fertilizer irrigation plan push module is used to select an irrigation plan based on all recommendation confidence levels and push it to the farmer's terminal when the farmer initiates a water and fertilizer irrigation suggestion request through the terminal.

[0058] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0059] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0060] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A water and fertilizer irrigation management method based on the integration of agricultural big data and AI, characterized in that, The water and fertilizer irrigation management method includes the following steps: Acquire agricultural big data and construct a knowledge graph of water and fertilizer irrigation; Semantic embedding vectors of soil moisture changes between adjacent irrigation events are extracted from water and fertilizer irrigation events under multiple consecutive planting seasons. The knowledge graph is then incrementally updated based on the soil moisture state transitions according to all the semantic embedding vectors to obtain the updated knowledge graph. Real-time soil data of field crops under the current planting season is collected and uploaded to the cloud. The water and fertilizer stress of field crops is analyzed by the artificial intelligence model in the cloud and the soil data to obtain the water and fertilizer stress level of field crops under the current planting season. Using the crop type identifier and growth stage identifier of the current planting season as the retrieval index, information subgraphs are extracted from the updated knowledge graph to obtain water and fertilizer irrigation information subgraphs associated with the current planting season. The weighted incentive coefficients are determined based on the stress interval in which the water and fertilizer stress levels are located; Take out an irrigation scheme entity node from the water and fertilizer irrigation information subgraph in sequence, and read the expected soil moisture response path stored in the irrigation scheme entity node. The expected soil moisture response path contains multiple expected soil moisture status nodes arranged in chronological order. Extract the current soil moisture value sequence from the real-time collected current soil moisture status, and align the current soil moisture value sequence with multiple expected soil moisture status nodes in the expected soil moisture response path according to time. Each moisture value in the current moisture value sequence after alignment is compared with the moisture value in the expected moisture state node at the corresponding time position. The number of matching pairs with differences within the preset tolerance range in all comparison results is counted. The ratio of the number of matching pairs to the total number of comparisons is used as the path fit. The original push weights of the irrigation scheme entity nodes are adjusted in combination based on the path matching degree and the weight incentive coefficient to obtain the updated push weights of the irrigation scheme entity nodes. A recommendation confidence distribution is constructed using the updated push weights of all irrigation scheme entity nodes, and then the recommendation confidence of each irrigation scheme under the current planting season is determined based on the recommendation confidence distribution. When farmers initiate a request for water and fertilizer irrigation suggestions through their terminals, an irrigation plan is selected based on the confidence level of all recommendations and pushed to the farmer's terminal.

2. The water and fertilizer irrigation management method based on the integration of agricultural big data and AI as described in claim 1, characterized in that, Acquiring agricultural big data and constructing a knowledge graph for water, fertilizer, and irrigation specifically includes: Acquire agricultural big data, which includes crop variety data, soil type data, historical meteorological data, historical irrigation data, and historical fertilization data; Entity extraction and relation extraction are performed on the agricultural big data to obtain multiple entities related to water and fertilizer irrigation and the relationships between these entities. The entities include crops, soil types, irrigation methods, fertilizer types, irrigation time nodes, and corresponding water and fertilizer ratios. Construct the initial structure of the knowledge graph for water and fertilizer irrigation based on all entities and the relationships between them; The irrigation and fertilization history records from the agricultural big data are used as attribute information to fill the corresponding entity nodes in the initial structure of the knowledge graph, thus obtaining the knowledge graph of water and fertilizer irrigation.

3. The water and fertilizer irrigation management method based on the integration of agricultural big data and AI as described in claim 1, characterized in that, The semantic embedding vectors for soil moisture changes between adjacent irrigation events, extracted from water and fertilizer irrigation events over multiple consecutive planting seasons, specifically include: Water and fertilizer irrigation events for multiple consecutive planting seasons are obtained from historical databases. Each water and fertilizer irrigation event records the irrigation time, irrigation amount, fertilizer amount, and soil moisture parameters before and after irrigation. The soil moisture parameters after the end of the first irrigation event in two adjacent irrigation events are taken as the initial moisture state, and the soil moisture parameters before the start of the second irrigation event are taken as the ending moisture state. Calculate the change in moisture parameters between the initial moisture state and the final moisture state to obtain the moisture change vector between two adjacent water and fertilizer irrigation events; The water and fertilizer change vector between two adjacent water and fertilizer irrigation events is determined based on the irrigation time, irrigation amount, and fertilizer application amount of the previous and subsequent water and fertilizer irrigation events. The initial soil moisture state, the final soil moisture state, the soil moisture change vector, and the water and fertilizer change vector are semantically encoded to generate a semantic embedding vector of soil moisture change between adjacent irrigation events.

4. The water and fertilizer irrigation management method based on the integration of agricultural big data and AI as described in claim 1, characterized in that, The knowledge graph is incrementally updated based on all semantic embedding vectors to reflect moisture state transitions, resulting in an updated knowledge graph that specifically includes: Temporal parsing is performed on each semantic embedding vector to extract the soil moisture baseline state before irrigation, the soil moisture recovery state after irrigation, and the soil moisture evolution trajectory between the two. Based on the baseline moisture state and the moisture recovery state, node anchoring is performed in the knowledge graph to locate the corresponding moisture state transition path node pairs; The soil moisture evolution trajectory is encoded as a transition edge attribute and mapped to the nodes of the soil moisture state transition path to form an incremental edge for soil moisture state transition; Traverse all semantic embedding vectors and integrate the incremental edges parsed from each semantic embedding vector into the knowledge graph in the order of planting seasons to construct a cross-seasonal soil moisture state transition chain. The connectivity of the transition paths of the fused knowledge graph is verified based on the moisture state transition chain. Multiple incremental edges with the same start node and end node are semantically aggregated to obtain the updated knowledge graph.

5. The water and fertilizer irrigation management method based on the integration of agricultural big data and AI as described in claim 4, characterized in that, The connectivity of the transition paths in the fused knowledge graph is verified based on the moisture state transition chain. Multiple incremental edges with the same start and end nodes are semantically aggregated to obtain the updated knowledge graph, which specifically includes: Using each transition node in the soil moisture state transition chain as the starting point of the traversal, the traversal is performed along the directed edges in the fused knowledge graph to mark all connected paths that start from the starting node and can reach the ending node. Filter out path groups with the same start node and the same end node from all connected paths, and extract the soil moisture evolution trajectory code carried by each incremental edge in each path group. All soil moisture evolution trajectory codes in each path are concatenated in chronological order to form a composite trajectory code. This composite trajectory code is then used to replace multiple incremental edges in that path, forming a single aggregated edge, thus obtaining the updated knowledge graph.

6. The water and fertilizer irrigation management method based on the integration of agricultural big data and AI as described in claim 1, characterized in that, By using cloud-based artificial intelligence models and soil data to analyze water and fertilizer stress in field crops, the specific water and fertilizer stress levels of field crops in the current planting season are obtained, including: The artificial intelligence model that has undergone transfer learning in advance is retrieved from the cloud, wherein the artificial intelligence model is a water and fertilizer stress recognition model; Determine the crop water tolerance threshold curve and crop nutrient tolerance threshold curve that match the crop variety and current planting stage in the field; The soil data is input into the water and fertilizer stress identification model. The water and nutrient deviation of the current soil moisture is determined by the identification results of the water and fertilizer stress identification model, the crop water tolerance threshold curve, and the crop nutrient tolerance threshold curve. The stress level is mapped based on the moisture deviation and the nutrient deviation, and the stress interval that the moisture deviation and the nutrient deviation fall into is determined as the water and fertilizer stress level of the field crop in the current planting season.

7. The water and fertilizer irrigation management method based on the integration of agricultural big data and AI as described in claim 6, characterized in that, The soil data is input into the water and fertilizer stress identification model. The determination of the current soil moisture deviation and nutrient deviation based on the identification results of the water and fertilizer stress identification model, the crop water tolerance threshold curve, and the crop nutrient tolerance threshold curve specifically includes: The soil data is converted into a soil feature vector, and the soil feature vector is input into the water and fertilizer stress identification model. The water and fertilizer stress confidence level output by the water and fertilizer stress identification model and the crop water tolerance threshold curve are used to perform water deviation analysis on the current soil moisture to obtain the water deviation degree of the current soil moisture. Nutrient deviation analysis of the current soil moisture is performed using the water and fertilizer stress confidence score output by the water and fertilizer stress identification model and the crop nutrient tolerance threshold curve to obtain the nutrient deviation degree of the current soil moisture.

8. A water and fertilizer irrigation management system based on the integration of agricultural big data and AI, used to execute the water and fertilizer irrigation management method based on the integration of agricultural big data and AI as described in any one of claims 1 to 7, characterized in that, The water and fertilizer irrigation management system includes: The knowledge graph construction module is used to acquire agricultural big data and construct a knowledge graph for water and fertilizer irrigation. The soil moisture semantic vector extraction and knowledge graph incremental update module is used to extract the semantic embedding vector of soil moisture changes between adjacent irrigation events from water and fertilizer irrigation events under multiple consecutive planting seasons, and to perform incremental updates of the knowledge graph based on all the semantic embedding vectors to obtain the updated knowledge graph. The crop water and fertilizer stress level analysis module is used to collect soil data of field crops in real time during the current planting season and upload it to the cloud. The cloud-based artificial intelligence model and the soil data are used to analyze the water and fertilizer stress of field crops and obtain the water and fertilizer stress level of field crops in the current planting season. The irrigation scheme confidence dynamic calibration module is used to push the weight update of the water and fertilizer irrigation information subgraph associated with the current planting season in the updated knowledge graph based on the water and fertilizer stress level, thereby determining the recommended confidence of each irrigation scheme under the current planting season. The precise push module for water and fertilizer irrigation plans is used to select an irrigation plan based on all recommendation confidence levels and push it to the farmer's terminal when the farmer initiates a water and fertilizer irrigation suggestion request through the terminal.

Images