Vegetable growth cycle pest control decision-making system and method based on mapping knowledge domain

Abstract

The invention relates to the technical field of agricultural informatization and intelligent decision making, and discloses a vegetable growth cycle pest control decision making system and method based on a knowledge graph. According to the method, an unstructured agricultural data stream including vegetable varieties, growth stages, environmental parameters, pest and disease records and geographic information is collected from an original database. A data tuple set with a time stamp is generated by performing scanning identification and content extraction on a data stream. Then, clustering is carried out according to the growth stages to which the tuples belong, and stage feature clusters sorted according to the growth cycle are formed; and performing multi-dimensional feature correlation analysis on the stage feature cluster, and establishing a dynamic coupling relationship among the environmental parameters, the disease and pest history and the geographic space features under the growth cycle dimension. And based on the dynamic coupling relationship, generating disease and pest control decision parameters of the vegetables in the specific planting area in each growth cycle stage. According to the invention, accurate matching between the control decision and the crop physiological time sequence and regional characteristics is realized.

Description

Technical Field

[0001] This invention relates to the field of agricultural informatization and intelligent decision-making technology, specifically to a knowledge graph-based decision-making system and method for the prevention and control of diseases and pests in the vegetable growth cycle. Background Technology

[0002] In current agricultural production management, pest and disease control decisions largely rely on the experience of agricultural technicians or rule bases based on static historical data. Common technical methods include establishing pest and disease databases and querying control solutions through keyword matching; or using environmental sensor data to set fixed thresholds to trigger alarms. These methods treat the entire growing season of vegetables as a homogeneous period or store and analyze data from different sources in isolation. Existing solutions deal with static or simply time-series data records, failing to structure and reorganize the data according to the inherent rhythms of vegetable physiological development. A disconnect exists between data records and the actual phenological stages of the crop, resulting in poor timeliness of control recommendations.

[0003] The shortcomings of existing technologies lie in their inability to reflect the dynamic interaction between environmental parameters, pest and disease occurrence patterns, and geographical conditions at different crop growth stages in their decision-making models. The correlation between environmental data and pest and disease records is mostly post-hoc, lacking forward-looking coupling analysis based on the growth cycle. Geospatial information is often merely used as background labels, failing to be deeply integrated into decision factors. This results in control strategies that are general in nature but weak in specificity, making it difficult to adapt to the specific contexts of specific regions and varieties at different growth stages, and thus unable to achieve truly precise control. This invention aims to address two core issues: how to ensure that data organization is strictly aligned with the growth timeline, and how to construct a dynamic correlation model of multi-source heterogeneous features within this timeline framework. Summary of the Invention

[0004] The purpose of this invention is to provide a knowledge graph-based decision-making system and method for the prevention and control of diseases and pests in the vegetable growth cycle, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a knowledge graph-based decision-making method for pest and disease control during the vegetable growth cycle, the method comprising:

[0006] Unstructured agricultural data streams containing vegetable varieties, growth stages, environmental parameters, pest and disease records, and geographic information were collected from the original database.

[0007] The unstructured agricultural data stream is scanned line by line to identify the segments in the data stream that are associated with the vegetable growth cycle, and the content of each segment is extracted to generate a set of data tuples containing time stamps;

[0008] The data tuple set is input into the feature fusion channel, and stage clustering is performed on the data tuples according to the vegetable growth stage to which the segment unit belongs, generating stage feature clusters sorted by growth cycle.

[0009] Multi-dimensional feature association analysis was performed on the aforementioned stage feature clusters to establish a dynamic coupling relationship between environmental parameter features, pest and disease occurrence history features, and geospatial features under the growth cycle dimension.

[0010] Based on the dynamic coupling relationship, decision parameters for pest and disease control of vegetables in a specific planting area at each stage of the growth cycle are generated.

[0011] Preferably, the unstructured agricultural data stream is scanned line by line to identify segments in the data stream that are associated with the vegetable growth cycle, and the content of each segment is extracted to generate a set of data tuples containing time stamps, including:

[0012] Receive real-time or historical unstructured agricultural data streams output from the raw database through a data interface;

[0013] A sliding analysis window is set on the unstructured agricultural data stream, and keyword matching is performed on the data content passing through the sliding analysis window. The keywords include feature descriptors of vegetable growth cycle stages.

[0014] When the data content in the sliding analysis window successfully matches any feature descriptor, the data content before and after the matching position with a preset length is extracted as a complete segment unit.

[0015] Entity and attribute extraction is performed on each segment unit to identify the vegetable variety name, specific growth stage, pest and disease type, occurrence time and geographical location description text, and the extraction results are encapsulated into a structured data tuple.

[0016] Each structured data tuple is added with a timestamp indicating its original appearance in the unstructured agricultural data stream, ultimately generating a set of data tuples containing the timestamps.

[0017] Preferably, the data tuple set is input into the feature fusion channel, and stage clustering is performed on the data tuples according to the vegetable growth stage to which the segment unit belongs, generating stage feature clusters sorted by growth cycle, including:

[0018] Establish a feature fusion channel, which contains multiple parallel feature processing branches;

[0019] Read the specific growth stage marked for each data tuple from the set of data tuples containing time stamps;

[0020] Based on the specific growth stage, the data tuples are routed to the corresponding feature processing branches.

[0021] Within each feature processing branch, data tuples belonging to the same specific growth stage are rearranged according to the order of their time stamps.

[0022] For the rearranged data tuples, feature vectorization transformation is performed based on the pest and disease type and environmental parameter category represented internally, and vectorization results with similar features within the same time period are aggregated into a feature subset;

[0023] The feature subsets output from all feature processing branches are assembled according to the natural order of their corresponding specific growth stages in the entire growth cycle of vegetables, generating stage feature clusters sorted by growth cycle.

[0024] Preferably, a multi-dimensional feature association analysis is performed on the aforementioned stage feature clusters to establish a dynamic coupling relationship between environmental parameter features, pest and disease occurrence history features, and geospatial features under the growth cycle dimension, including:

[0025] The stage feature clusters ordered by the growth cycle are analyzed to separate environmental parameter feature subsequences, pest and disease occurrence history feature subsequences, and geospatial feature subsequences.

[0026] For each specific growth stage, create a dynamic association graph model corresponding to that growth stage;

[0027] In the dynamic association graph model, the feature values ​​of the environmental parameter feature subsequence in the growth stage are used as environmental nodes, the feature values ​​of the pest and disease occurrence history feature subsequence in the growth stage are used as pest and disease nodes, and the feature values ​​of the geospatial feature subsequence in the growth stage are used as spatial nodes.

[0028] By analyzing the co-occurrence and transfer patterns among environmental nodes, pest and disease nodes, and spatial nodes in historical data, directed connection edges among these nodes are defined and created.

[0029] Each of the directed connection edges is accompanied by an association strength weight calculated from historical statistical frequencies, thereby forming a dynamic coupled subgraph describing the multidimensional feature interaction relationships within the growth stage;

[0030] The dynamic coupling subgraphs corresponding to each growth stage are merged according to the growth cycle order to generate dynamic coupling relationships that run through the entire growth cycle.

[0031] The analysis of the stage feature clusters ordered by growth cycle is used to separate environmental parameter feature subsequences, pest and disease occurrence history feature subsequences, and geospatial feature subsequences, including:

[0032] Load the stage feature clusters sorted by growth cycle, and deconstruct the feature subsets corresponding to each growth stage in the stage feature clusters;

[0033] From each feature subset, data labeled with feature vectors such as temperature, humidity, light intensity, and soil pH are identified and extracted, and arranged in chronological order to form the environmental parameter feature fragments of the growth stage.

[0034] From the same feature subset, identify and extract data with feature vector labels as pest and disease name, occurrence time, and severity, and arrange them in chronological order to form a historical feature segment of pest and disease occurrence in the growth stage.

[0035] From the same feature subset, identify and extract data whose feature vector labels are latitude and longitude coordinates, altitude, and plot number, and arrange them in chronological order to form the geospatial feature fragments of the growth stage.

[0036] The environmental parameter feature segments, pest and disease occurrence history feature segments, and geospatial feature segments corresponding to each growth stage are vertically concatenated according to their growth stage order to generate environmental parameter feature subsequences, pest and disease occurrence history feature subsequences, and geospatial feature subsequences.

[0037] Preferably, by analyzing the co-occurrence and transfer patterns among the environmental nodes, pest and disease nodes, and spatial nodes in historical data, directed connection edges are defined and created among the environmental nodes, pest and disease nodes, and spatial nodes, including:

[0038] Collect records of the environmental parameter feature subsequence, the pest and disease occurrence history feature subsequence, and the geospatial feature subsequence in multiple historical growth cycles;

[0039] The co-occurrence frequency is calculated as the number of times that a specific environmental parameter characteristic node, a specific pest and disease occurrence history characteristic node, and a specific geospatial characteristic node occur simultaneously within the same growth stage.

[0040] The frequency of transitions is calculated as the number of times a specific environmental parameter characteristic node transitions to a specific historical characteristic node of pest and disease occurrence between adjacent growth stages, and the number of times a specific geographic spatial characteristic node transitions to a specific historical characteristic node of pest and disease occurrence.

[0041] Set a co-occurrence frequency threshold and a transfer frequency threshold. When the co-occurrence frequency exceeds the threshold, create undirected connection edges between each pair of the environmental node, pest node, and spatial node to indicate the existence of an association.

[0042] When the frequency of the transition exceeds its threshold, a directed connection edge representing a causal or influence relationship is created from the node in the starting state to the node in the arriving state.

[0043] The directed and undirected edges are merged to form the set of connecting edges between nodes in the growth stage.

[0044] Preferably, the dynamic coupling subgraphs corresponding to each growth stage are fused according to the growth cycle sequence to generate dynamic coupling relationships that span the entire growth cycle, including:

[0045] Obtain the dynamic coupling subgraph corresponding to the first growth stage and use it as the initial fusion graph;

[0046] Read the dynamic coupling subgraph corresponding to the next growth stage in the order of the growth cycle.

[0047] Identify the same nodes in the initial fusion graph and the dynamic coupling subgraph of the next growth stage, where the same nodes refer to the same geospatial features or the same vegetable varieties;

[0048] Merge the nodes that are the same as those in the initial fusion graph in the next growth stage dynamic coupling subgraph.

[0049] In the next growth stage dynamic coupling subgraph, unique new nodes and new connecting edges are added to the initial fusion graph;

[0050] In the merged graph, between nodes of adjacent growth stages, temporal connection edges representing the cross-stage evolution of pests and diseases are added based on the sub-sequence of the historical characteristics of the occurrence of pests and diseases.

[0051] Repeat the process until all the dynamic coupling subgraphs of all growth stages are merged, ultimately generating a dynamic coupling relationship graph that integrates all feature nodes and complex connections throughout the entire growth cycle.

[0052] Preferably, based on the dynamic coupling relationship, decision parameters for pest and disease control of vegetables in a specific planting area at each stage of the growth cycle are generated, including:

[0053] Receive a query request for a target vegetable variety in a specific growing area, the query request containing a geolocation code;

[0054] Traverse the dynamic coupling relationship, filter out the geospatial feature nodes that match the geographic location code, and locate the growth stage to which the geospatial feature node belongs;

[0055] Extract the dynamic coupling subgraph corresponding to the growth stage, and obtain all environmental parameter feature nodes and pest and disease occurrence history feature nodes that have directed connection edges with the geospatial feature nodes;

[0056] Read the current monitoring value of the environmental parameter feature node, compare it with the historical feature value range of the environmental parameter feature node recorded in the dynamic coupling subgraph, and calculate the deviation.

[0057] By combining the association strength weights on the directed connection edges, the historical feature nodes of pest and disease occurrence associated with the current environmental monitoring value are weighted and evaluated to predict the probability level of pest and disease occurrence.

[0058] Based on the probability level of the occurrence of pests and diseases and the associated pest and disease types, pest and disease control decision parameters including the timing of prevention and control, recommended measures and intensity levels are generated from a preset strategy library.

[0059] Preferably, the current monitoring value of the environmental parameter feature node is read, and compared with the historical feature value range of the environmental parameter feature node recorded in the dynamic coupling subgraph to calculate the deviation, including:

[0060] Obtain the actual monitored values ​​of environmental parameters for the specific planting area at the current growth stage from a real-time data source;

[0061] From the dynamic coupling subgraph, locate the geospatial feature node corresponding to the geographic location code, and find the environmental parameter feature node directly connected to it;

[0062] Query the feature values ​​of the environmental parameter feature nodes recorded in historical data. The feature values ​​include the maximum value, minimum value, average value, and common numerical distribution ranges.

[0063] The actual monitored values ​​of the environmental parameters are compared with the historical maximum, historical minimum, and historical average values, and the absolute difference is calculated.

[0064] Divide the absolute difference by the corresponding historical feature value change range to obtain multiple relative deviation ratios for the maximum, minimum, and average values;

[0065] The maximum value among the multiple relative deviation ratios is selected as the final deviation of the environmental parameter feature node at the current moment.

[0066] Preferably, by combining the association strength weights on the directed connection edges, a weighted evaluation is performed on the historical characteristic nodes of pest and disease occurrence associated with the current environmental monitoring value to predict the probability level of pest and disease occurrence, including:

[0067] In the dynamic coupling subgraph, find all directed connection edges that start from the environmental parameter feature nodes and point to different pest and disease occurrence history feature nodes.

[0068] Obtain the association strength weight attached to each of the directed connection edges, the association strength weight being calculated based on historical co-occurrence and transition frequency;

[0069] The final deviation of the calculated environmental parameter feature nodes is used as an influencing factor;

[0070] The historical occurrence frequency of each disease and pest occurrence node pointed to by each directed connection edge is used as another influencing factor.

[0071] Multiply the correlation strength weight, the final deviation influence factor, and the historical occurrence frequency influence factor to obtain a comprehensive occurrence index for each possible historical characteristic node of pest and disease occurrence;

[0072] Sort all historical characteristic nodes of pest and disease occurrence from high to low according to the comprehensive occurrence index;

[0073] Based on a preset index threshold range, the comprehensive occurrence index is mapped to different level labels, including high probability, medium probability, and low probability, thereby generating a pest and disease occurrence probability level.

[0074] Preferably, when the processor executes the computer program, it implements the steps of the knowledge graph-based decision-making method for pest and disease control in the vegetable growth cycle as described in any of the above-mentioned methods.

[0075] Compared with the prior art, the beneficial effects of the present invention are:

[0076] Data tuples are clustered according to their specific growth stages based on their time stamps, generating stage feature clusters arranged sequentially according to the growth cycle. This transforms the original data stream, simply listed by collection time, into a data structure that strictly corresponds to the physiological development stages of vegetables. Each stage's data forms an independent and ordered cluster, achieving precise alignment between data and the crop life cycle. This eliminates analytical biases caused by temporal misalignment, providing each independent growth stage with a clearly defined and pure data set. This allows subsequent analysis to be based on the unique physiological state and vulnerability of each crop stage, supporting a phased and differentiated decision-making process, rather than providing a general solution for the entire growing season.

[0077] Based on feature clusters organized by growth stages, a dynamic coupling relationship is established between environmental parameters, pest and disease history, and geospatial features along the growth cycle dimension. Instead of simple feature splicing or static association, it dynamically models the interaction patterns of three types of heterogeneous features within each growth stage as the analysis unit. Geospatial features are integrated into the association network as core variables rather than background information. This generates a feature association map that evolves with the growth stage and incorporates spatial context. The coupling relationship can explain why the risk of pest and disease occurrence differs in different regions or at different stages under the same environmental conditions, thus directly driving the generation of highly contextualized prevention and control parameters. These parameters are essentially the output of the dynamic model under the current stage and regional conditions, exhibiting stronger interpretability and regional adaptability. Attached Figure Description

[0078] Figure 1 This is a schematic diagram illustrating the working principle of the knowledge graph-based decision-making method for pest and disease control during the vegetable growth cycle as described in this invention.

[0079] Figure 2 A flowchart for generating stage feature clusters;

[0080] Figure 3 A flowchart for defining and creating directed connection edges;

[0081] Figure 4 A bar chart showing the correlation between environmental parameters and the strength of the relationship between pests and diseases;

[0082] Figure 5 This is a bar chart comparing the frequency of pest and disease occurrence with the overall occurrence index. Detailed Implementation

[0083] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0084] Please see Figure 1This invention provides a knowledge graph-based decision-making method for pest and disease control in vegetable growth cycles. The method includes: collecting unstructured agricultural data streams containing vegetable varieties, growth stages, environmental parameters, pest and disease records, and geographic information from an original database; scanning the unstructured agricultural data streams one by one and identifying segment units in the data streams that are associated with the vegetable growth cycle; extracting the content of each segment unit to generate a set of data tuples containing time stamps; inputting the data tuple set into a feature fusion channel and performing stage clustering operations on the data tuples according to the vegetable growth stage to which the segment unit belongs, generating stage feature clusters sorted by growth cycle; performing multi-dimensional feature association analysis on the stage feature clusters to establish a dynamic coupling relationship between environmental parameter features, historical features of pest and disease occurrence, and geographic spatial features under the growth cycle dimension; and generating pest and disease control decision parameters for vegetables in a specific planting area at each growth cycle stage based on the dynamic coupling relationship.

[0085] In one embodiment of the present invention, see [reference] Figure 2 The system receives real-time or historical unstructured agricultural data streams from the original database via a data interface. A sliding analysis window is set up on the unstructured agricultural data stream, and keyword matching is performed on the data content passing through the sliding analysis window. Keywords include feature descriptors of vegetable growth cycle stages. When the data content within the sliding analysis window successfully matches any feature descriptor, the data content before and after the matching position of a preset length is extracted as a complete segment unit. Entity and attribute extraction is performed on each segment unit, and the vegetable variety name, specific growth stage, pest and disease type, occurrence time, and geographical location description text are identified. The extraction results are encapsulated into a structured data tuple. A timestamp is added to each structured data tuple to indicate its original appearance in the unstructured agricultural data stream, and finally, a set of data tuples containing timestamps is generated. A feature fusion channel is established, which contains multiple parallel feature processing branches. The specific growth stage labeled for each data tuple is read from the set of data tuples containing time stamps. The data tuples are routed to the corresponding feature processing branches according to the specific growth stage. Within each feature processing branch, data tuples belonging to the same specific growth stage are rearranged according to the order of their time stamps. The rearranged data tuples are transformed into feature vectors based on the pest and disease type and environmental parameter category represented by them. The vectorized results with similar features within the same time period are aggregated into a feature subset. The feature subsets output by all feature processing branches are assembled according to the natural order of their corresponding specific growth stages in the whole growth cycle of vegetables to generate stage feature clusters sorted by growth cycle.

[0086] In practice, the system receives real-time unstructured agricultural data streams from the raw database of the vegetable planting management system via an application programming interface (API). This unstructured data stream includes temperature values ​​recorded by sensors, pest and disease observation logs entered by agricultural personnel, geographic location coordinates obtained from satellite remote sensing, and growth stage descriptions from planting records. A 500-character sliding analysis window is set up on the unstructured agricultural data stream. The window moves forward in 100-character increments, matching each data entry through the window with a preset keyword list. This list includes descriptors representing vegetable growth cycle stages such as "sowing period," "seedling stage," "flowering stage," and "fruiting stage." When a data entry within the sliding analysis window successfully matches the descriptor "seedling stage," the system extracts 300 characters of data before and after the match, forming a complete unit of information about seedling growth.

[0087] For each segment unit, entity and attribute extraction operations based on a named entity recognition model are performed to identify and extract the vegetable variety name "tomato", the specific growth stage "seedling stage", the pest type "aphid", the occurrence time "2023-05-10", and the geographical location description text "40 degrees north latitude, 116 degrees east longitude". The extraction results are encapsulated into a structured data tuple with uniform fields. A timestamp "1683724800" corresponding to its original appearance in the unstructured agricultural data stream is added to each structured data tuple, ultimately generating a set of data tuples containing timestamps. In some embodiments, the original database also outputs historical unstructured agricultural data streams. The same sliding analysis window and feature descriptor matching mechanism are used to process the historical unstructured agricultural data streams. Segment units in the historical unstructured agricultural data streams are also extracted and have their corresponding timestamps added, collectively forming a set of data tuples.

[0088] A feature fusion channel is established, containing multiple parallel feature processing branches corresponding to the sowing, seedling, flowering, and fruiting stages, respectively. From a set of data tuples containing timestamps, the specific growth stage field content labeled for each data tuple is read. Based on the specific growth stage field content, the data tuples are routed to the "sowing stage," "seedling stage," "flowering stage," or "fruiting stage" feature processing branches, respectively. Within each feature processing branch, data tuples belonging to the same specific growth stage are rearranged in ascending order according to the value of their timestamps. Optionally, the timestamp format is a Unix timestamp.

[0089] The rearranged data tuples are transformed into feature vectors based on the pest and disease types and environmental parameter categories they represent. In specific implementations, one-hot encoding is used to convert pest and disease types such as "aphids" and "powdery mildew," as well as environmental parameter categories such as "temperature" and "humidity," into fixed-dimensional sparse vectors. In further implementations, vectors with a cosine similarity higher than a set threshold in the feature vectorization results within the same natural day are aggregated into a feature subset. The feature subsets output from all feature processing branches are then assembled according to the natural order of their corresponding specific growth stages in the tomato's entire growth cycle—namely, sowing period, seedling stage, flowering period, and fruiting period—to generate stage feature clusters ordered by growth cycle. In some embodiments, the number of feature processing branches is strictly consistent with the number of standard growth stages defined for the target vegetable variety.

[0090] Optionally, the length and step size of the sliding analysis window can be preset and adjusted based on the average length of statements in the unstructured agricultural data stream. It can be understood that the feature descriptor list for vegetable growth cycle stages can be expanded and maintained based on agronomical knowledge of different vegetable varieties. It can also be understood that during feature vectorization, specific values ​​of environmental parameters, such as a temperature of 25 degrees Celsius, can be normalized to convert them into scalar values ​​before being used as a dimension of the vector. The criteria for judging feature similarity can be achieved by calculating the Euclidean distance or cosine similarity between vectors and comparing it with a preset threshold. The feature subsets output from all feature processing branches are assembled according to the natural order of their corresponding specific growth stages in the entire vegetable growth cycle. The assembly process can be represented as a sequential splicing operation, generating stage feature clusters sorted by growth cycle.

[0091] In one embodiment of the present invention, the stage feature clusters ordered by the growth cycle are analyzed and environmental parameter feature subsequences, pest and disease occurrence history feature subsequences, and geospatial feature subsequences are separated from them. For each specific growth stage, a dynamic association graph model corresponding to the growth stage is created. In the dynamic association graph model, the feature values ​​of the environmental parameter feature subsequences in the growth stage are used as environmental nodes, the feature values ​​of the pest and disease occurrence history feature subsequences in the growth stage are used as pest and disease nodes, and the feature values ​​of the geospatial feature subsequences in the growth stage are used as spatial nodes. By analyzing the co-occurrence and transfer patterns between environmental nodes, pest and disease nodes, and spatial nodes in historical data, directed connection edges between environmental nodes, pest and disease nodes, and spatial nodes are defined and created. Each directed connection edge is attached with an association strength weight calculated by historical statistical frequency, thereby forming a dynamic coupling subgraph describing the multidimensional feature interaction relationship within the growth stage. The dynamic coupling subgraphs corresponding to each growth stage are fused according to the growth cycle order to generate a dynamic coupling relationship that runs through the entire growth cycle. The system loads feature clusters ordered by growth cycle and deconstructs the feature subsets corresponding to each growth stage within each feature cluster. It identifies and extracts data with feature vector labels of temperature, humidity, light intensity, and soil pH from each feature subset and arranges them chronologically to form environmental parameter feature segments for each growth stage. It also identifies and extracts data with feature vector labels of pest / disease name, occurrence time, and severity from the same feature subset and arranges them chronologically to form historical pest / disease occurrence feature segments for each growth stage. Finally, it identifies and extracts data with feature vector labels of latitude / longitude coordinates, altitude, and plot number from the same feature subset and arranges them chronologically to form geospatial feature segments for each growth stage. The system then vertically concatenates the environmental parameter feature segments, historical pest / disease occurrence feature segments, and geospatial feature segments corresponding to each growth stage according to their growth stage order to generate environmental parameter feature subsequences, historical pest / disease occurrence feature subsequences, and geospatial feature subsequences.

[0092] In specific implementation, the stage feature clusters generated by the feature fusion channel and sorted by growth cycle are loaded. The stage feature clusters include feature subsets for sowing period, seedling period, flowering period, and fruiting period. The feature subsets corresponding to each growth stage in the stage feature clusters are deconstructed. Data points with feature vector labels of temperature "22℃", "24℃", "20℃", humidity "65%RH", "70%RH", "68%RH", light "12000lux", "11500lux", and soil pH "pH6.5" "pH6.8" are identified and extracted from the sowing period feature subset. These data points are arranged in the order of their recorded time markers "t1", "t2", "t3" to form the environmental parameter feature fragments of the sowing period growth stage. In practice, from the same subset of sowing period features, data points labeled with the name of the pest or disease, such as "damping-off," with the feature vector label "t2" indicating the time of occurrence, and with the feature vector label "mild," indicating the severity, are identified and extracted. These data points are then arranged in chronological order to form a historical feature segment of pest and disease occurrence during the sowing period. Similarly, data points labeled with latitude and longitude coordinates, such as "40.1°N, 116.2°E," with the feature vector label "50 meters," indicating altitude, and with the feature vector label "B-05," indicating plot number, are identified and extracted from the sowing period feature subset. These data points are then arranged in chronological order to form a geospatial feature segment of the sowing period.

[0093] For each specific growth stage, such as the seedling stage, a dynamic correlation graph model is created. In this model, the characteristic value of the environmental parameter feature sequence "average daily temperature 25℃" during the seedling stage is used as an environmental node; the characteristic value of the historical pest and disease occurrence feature sequence "aphid outbreak" during the seedling stage is used as a pest and disease node; and the characteristic value of the geospatial feature sequence "plot B-05" during the seedling stage is used as a spatial node. By analyzing the co-occurrence and transfer patterns among environmental, pest and disease, and spatial nodes in historical data, directed connections are defined and created between these nodes. Each directed connection is accompanied by a correlation strength weight calculated from historical statistical frequencies, thus forming a dynamic coupled subgraph describing the multidimensional feature interactions within the seedling stage.

[0094] The environmental parameter feature segments, pest and disease occurrence history feature segments, and geospatial feature segments corresponding to each growth stage are vertically concatenated according to their growth stage sequence to generate environmental parameter feature subsequences, pest and disease occurrence history feature subsequences, and geospatial feature subsequences. In specific implementations, the environmental parameter feature segments of the sowing period, seedling period, flowering period, and fruiting period are vertically connected in the order of sowing period, seedling period, flowering period, and fruiting period to generate environmental parameter feature subsequences covering the entire cycle. The dynamic coupling subgraphs corresponding to each growth stage are fused according to the growth cycle sequence to generate dynamic coupling relationships that span the entire growth cycle. In some embodiments, the generated geospatial feature subsequence is a list of geographical location descriptions arranged by time stage. In some embodiments, the pest and disease occurrence history feature subsequence contains an ordered set of pest and disease types and their occurrence times recorded at different growth stages.

[0095] Optionally, feature vector label identification is based on precise matching of the metadata label field attached to each data point in the feature subset. It can be understood that the generation of environmental parameter feature subsequences is achieved by concatenating the environmental parameter feature fragments of each stage. It can also be understood that the node attribute values ​​in the dynamic association graph model directly originate from the feature values ​​of the corresponding feature subsequences at the corresponding growth stages. By analyzing the co-occurrence and transfer patterns among environmental nodes, pest nodes, and spatial nodes in historical data, a co-occurrence pattern refers to the pattern where multiple node feature values ​​appear simultaneously at the same time or in the same growth stage in historical records. The logic for creating directed edges is based on the chronological order of feature state changes in historical events. For example, if historical data repeatedly shows the appearance of an environmental node "temperature consistently above 28℃," followed by a pest node "red spider mite disease," then a directed edge is created from the former to the latter. The association strength weight can be calculated based on the proportion of co-occurrence or transfer events in the total historical events, with the association strength weight value ranging between 0 and 1.

[0096] In one embodiment of the present invention, see [reference] Figure 3This study collects records of environmental parameter feature subsequences, pest and disease occurrence history feature subsequences, and geospatial feature subsequences across multiple historical growth cycles. The co-occurrence frequency is calculated by counting the number of times the environmental parameter feature node's value range, the pest and disease occurrence history feature node's type, and the geospatial feature node's location occur simultaneously within the same growth stage. The transition frequency is calculated by counting the number of times the environmental parameter feature node transitions to the pest and disease occurrence history feature node's state and the number of times the geospatial feature node transitions to the pest and disease occurrence history feature node's state between adjacent growth stages. Co-occurrence frequency thresholds and transition frequency thresholds are set. When the co-occurrence frequency exceeds the threshold, undirected edges representing associations are created between each pair of environmental, pest and disease, and geospatial nodes. When the transition frequency exceeds the threshold, directed edges representing causal or influence relationships are created from the node in the initial state to the node in the final state. The directed and undirected edges are then merged into a set of connection edges between nodes within the growth stage. The process involves obtaining the dynamic coupling subgraph corresponding to the first growth stage and using it as the initial fusion graph. The dynamic coupling subgraph corresponding to the next growth stage is then read sequentially according to the growth cycle. Nodes identical to those in the initial fusion graph and the next growth stage's dynamic coupling subgraph are identified; these identical nodes represent the same geographic spatial features or the same vegetable variety. Nodes identical to those in the next growth stage's dynamic coupling subgraph and the initial fusion graph are merged. New nodes and new connecting edges unique to the next growth stage's dynamic coupling subgraph are added to the initial fusion graph. In the merged graph, temporal connecting edges representing the cross-stage evolution of pests and diseases are added between nodes in adjacent growth stages based on the historical feature subsequence of pest and disease occurrence. This process is repeated until all dynamic coupling subgraphs for all growth stages are fused, ultimately generating a dynamic coupling relationship graph that integrates all feature nodes and complex connections throughout the entire growth cycle.

[0097] In practice, records of environmental parameter characteristic subsequences, pest and disease occurrence history characteristic subsequences, and geospatial characteristic subsequences were collected across multiple historical growth cycles. These historical records covered the complete three-year planting process of the same vegetable variety in the same geographical area. The frequency of simultaneous occurrence of the environmental parameter characteristic node "daily average temperature between 25℃ and 28℃," the pest and disease occurrence history characteristic node "aphids," and the geospatial characteristic node "plot B-05" within the same seedling growth stage was recorded as the co-occurrence frequency. The frequency of transition from the environmental parameter characteristic node "air humidity consistently above 85%" to the pest and disease occurrence history characteristic node "downy mildew" between adjacent seedling and flowering growth stages was also recorded as the transition frequency. The frequency of transition from the geospatial characteristic node "low-lying area" to the pest and disease occurrence history characteristic node "root rot" was also recorded as the transition frequency. Co-occurrence frequency thresholds and transition frequency thresholds were set: a co-occurrence frequency threshold of 5 times and a transition frequency threshold of 3 times. If the co-occurrence frequency of "daily average temperature between 25℃ and 28℃", "aphids", and "plot B-05" exceeds the co-occurrence frequency threshold of 5 times, then an undirected connection edge representing the association is created between each pair of the environmental node "daily average temperature between 25℃ and 28℃", the "aphid" pest node, and the "plot B-05" spatial node. If the transfer frequency from the environmental node "air humidity consistently above 85%" to the "downy mildew" pest node exceeds the transfer frequency threshold of 3 times, then a directed connection edge representing a causal or influence relationship is created from the environmental node "air humidity consistently above 85%" to the "downy mildew" pest node. The created directed and undirected connections are merged to form the set of connections between nodes during the seedling growth stage.

[0098] Obtain the dynamic coupling subgraph corresponding to the first growth stage, i.e., the sowing stage, and use it as the initial fusion graph. Following the growth cycle sequence, read the dynamic coupling subgraph corresponding to the next growth stage, i.e., the seedling stage. Identify nodes identical to those in the initial fusion graph and the seedling stage dynamic coupling subgraph. These identical nodes represent the same geospatial feature "Plot B-05" or the same vegetable variety "Tomato". Merge the nodes representing the same geospatial feature "Plot B-05" in the seedling stage dynamic coupling subgraph and the initial fusion graph. This merging operation means that only one node representing "Plot B-05" is retained in the fused graph. Add new nodes unique to the seedling stage dynamic coupling subgraph, such as the seedling-specific environmental parameter feature node "sunshine duration exceeds 12 hours," and new connecting edges linking these new nodes to other nodes, to the initial fusion graph. In the merged graph, between the historical feature node of pest and disease occurrence "damping-off" during the sowing stage and the historical feature node of pest and disease occurrence "sudden drop" during the seedling stage, a temporal connection edge representing the cross-stage evolution of pests and diseases is added based on the disease evolution path recorded in the subsequence of historical feature nodes of pest and disease occurrence. The process of reading the dynamic coupling subgraph of the next growth stage in the order of the growth cycle, identifying and merging identical nodes, adding new nodes and new connections, and adding temporal connection edges is repeated until the dynamic coupling subgraphs corresponding to the flowering and fruiting stages are merged. Finally, a dynamic coupling relationship graph integrating all environmental parameter feature nodes, historical feature nodes of pest and disease occurrence, geospatial feature nodes, and complex connections throughout the entire tomato growth cycle is generated. In some embodiments, the specific values ​​of the co-occurrence frequency threshold and the transition frequency threshold are dynamically adjusted based on the total amount of historical data. In some embodiments, the direction of the temporal connection edge always points to a later growth stage.

[0099] Optionally, the association strength weight can be calculated based on historical statistical frequency, using the following formula:

[0100]

[0101] in: Represents the association strength weight. This represents the co-occurrence frequency or transition frequency obtained from historical data. This represents the total number of historical observations. It can be understood that the node merging operation involves integrating and aggregating the attribute information carried by each node before merging. It can also be understood that a dynamic coupling relationship graph is a knowledge representation stored in a graph structure, where nodes represent entities or states, and edges represent relationships or influences. In specific implementation, the total number of historical observations... This refers to the total number of times the target's growth stage or state transition is judged within a time window of historical data analysis. (Association strength weight) The value ranges from 0 to 1, with values ​​closer to 1 indicating stronger historical correlation. In practice, adding temporal connection edges is not limited to pest and disease nodes; significant changes in environmental parameter feature nodes between adjacent growth stages can also be expressed through temporal connection edges.

[0102] In one embodiment of the present invention, a query request for a target vegetable variety in a specific planting area is received, and the query request includes a geolocation code. The dynamic coupling relationship is traversed and geospatial feature nodes that match the geolocation code are selected and the growth stage to which the geospatial feature node belongs is located. The dynamic coupling subgraph corresponding to the growth stage is extracted and all environmental parameter feature nodes and pest and disease occurrence history feature nodes that have directed connection edges with the geospatial feature nodes are obtained. The current monitoring value of the environmental parameter feature node is read and compared with the range of historical feature values ​​of the environmental parameter feature node recorded in the dynamic coupling subgraph to calculate the deviation. The pest and disease occurrence history feature nodes associated with the current environmental monitoring value are weighted and evaluated in combination with the association strength weight on the directed connection edge to predict the pest and disease occurrence probability level. Based on the pest and disease occurrence probability level and the associated pest and disease type, pest and disease control decision parameters including control timing, recommended measures and intensity level are generated from a preset strategy library.

[0103] In practice, the system receives a query request for the target vegetable variety "tomato" in a specific planting area, containing the geolocation code "B-05". It then traverses the dynamic coupling graph, filtering out geospatial feature nodes that match the geolocation code "B-05" and locating the growth stage "seedling stage" to which the geospatial feature node "B-05" belongs. The system extracts the dynamic coupling subgraph corresponding to the seedling stage, obtaining all environmental parameter feature nodes and historical pest and disease occurrence feature nodes that have directed edges to the geospatial feature node "B-05". For example, it obtains the environmental parameter feature node "daily average temperature" and the historical pest and disease occurrence feature node "aphids". The system reads the current monitoring value of the environmental parameter feature node "daily average temperature" ("25℃") and compares it with the historical feature value range of "daily average temperature" recorded in the dynamic coupling subgraph ("20℃ to 30℃"), calculating the deviation of the current monitoring value from the historical range. By combining the association strength weights on directed edges, a weighted evaluation is performed on the "aphid" historical feature node associated with the current environmental monitoring value to predict the probability level of pest occurrence. Based on the predicted probability level "medium probability" and the associated pest type "aphid," pest control decision parameters including the control timing "pre-budding," the suggested measure "spraying imidacloprid," and the intensity level "standard dosage" are generated from a preset strategy library. In some embodiments, the query request can be entered through a graphical user interface, and the query request may include target growth stage information in addition to geolocation encoding. In some embodiments, nodes with directed edges to geospatial feature nodes in the dynamic coupling subgraph may include multiple environmental parameter feature nodes and multiple historical feature nodes of pest occurrence.

[0104] In practice, the calculation of deviation involves comparing the current monitored value with the historical characteristic value range. The process of reading the current monitored value of the environmental parameter characteristic node and comparing it with the historical characteristic value range of the environmental parameter characteristic node recorded in the dynamic coupling subgraph to calculate the deviation can be expressed as the formula:

[0105]

[0106] in: This represents the final deviation calculated. This represents the actual monitored values ​​of environmental parameters obtained from real-time data sources. This represents the historical maximum value of the environmental parameter feature node queried in the dynamic coupling subgraph. This represents the historical minimum value of the environmental parameter feature node queried in the dynamic coupling subgraph. This represents the historical average value of the environmental parameter feature nodes queried in the dynamic coupling subgraph. Represents the range of historical characteristic value changes, calculated as follows: In practice, the weighted evaluation process utilizes the weights of directed edges and the calculated deviation. The pre-defined strategy library is a database storing the mapping relationships between pest and disease types, occurrence probability levels, and recommended control parameters. Optionally, the occurrence probability levels of pests and diseases can be further refined into multiple discrete levels. It can be understood that the generation of pest and disease control decision parameters is the final output of the query request. The mapping process searches the pre-defined strategy library based on the pest and disease type and probability level as a joint primary key. The content of the pre-defined strategy library is predefined based on agricultural expert knowledge; see Table 1 for some mapping relationships.

[0107] Table 1: Mapping Table of Pest and Disease Control Strategies

[0108] Pests and diseases probability level Timing of prevention and control Recommended measures Intensity level aphid High probability Implement immediately Spraying flupyridine Increased dosage aphid Medium probability Before the buds appear Spray imidacloprid Standard dose Downy mildew Medium probability Early stage of the disease Spray dimethomorph Standard dose root rot Low probability After planting Apply hymexazol preventive dose

[0109] See Figure 4 This is a bar chart showing the correlation between environmental parameters and pests / diseases. Relative humidity has the highest correlation weight with downy mildew (approximately 0.9), making it the core environmental factor affecting this disease. Daily average temperature has a relatively high correlation weight with aphids (approximately 0.85), making it a major environmental driver of aphid occurrence. Soil pH has a higher correlation weight with root rot (0.8) than other diseases, reflecting the significant impact of soil acidity / alkalinity on root rot. Sunlight duration has a low correlation with all three types of pests / diseases, indicating it is a secondary influencing factor. This type of chart is commonly used for decision analysis in agricultural pest and disease control, helping growers to specifically monitor highly correlated environmental parameters. It is a visual representation of the dynamic coupling relationship between "environment" and "pests / diseases," providing data support for subsequent predictions of pest and disease occurrence probabilities.

[0110] In one embodiment of the present invention, the actual monitored values ​​of environmental parameters of a specific planting area at the current growth stage are obtained from a real-time data source. The geospatial feature node corresponding to the geographic location code is located from the dynamic coupling subgraph, and the environmental parameter feature node directly connected to it is found. The feature values ​​recorded by the environmental parameter feature node in historical data are queried, and the feature values ​​include the maximum value, the minimum value, the average value, and the common value distribution range. The actual monitored values ​​of environmental parameters are compared with the historical maximum value, the historical minimum value, and the historical average value to calculate the absolute difference. The absolute difference is divided by the corresponding historical feature value change range to obtain multiple relative deviation ratios for the maximum value, the minimum value, and the average value. The maximum value among the multiple relative deviation ratios is selected as the final deviation of the environmental parameter feature node at the current moment. In the dynamic coupling subgraph, all directed edges originating from environmental parameter feature nodes and pointing to different historical feature nodes of pest and disease occurrence are identified. The association strength weight attached to each directed edge is obtained, and the association strength weight is calculated based on historical co-occurrence and transfer frequency. The final deviation of the calculated environmental parameter feature nodes is used as an influencing factor, and the historical occurrence frequency of the historical feature nodes of pest and disease occurrence pointed to by each directed edge is used as another influencing factor. The association strength weight, the final deviation influencing factor, and the historical occurrence frequency influencing factor are multiplied to obtain the comprehensive occurrence index for each possible historical feature node of pest and disease occurrence. All historical feature nodes of pest and disease occurrence are sorted from high to low according to the comprehensive occurrence index. The comprehensive occurrence index is mapped to different level labels according to the preset index threshold range, and the level labels include high probability, medium probability, and low probability, thereby generating the pest and disease occurrence probability level.

[0111] In practice, the actual environmental parameters of the specific planting area "Plot B-05" during the current seedling growth stage are obtained from real-time data sources. These parameters include "daily average temperature 26.5℃" and "relative humidity 82%". From the dynamic coupling sub-graph corresponding to the seedling growth stage, the geospatial feature node "B-05" corresponding to the geographic location code "B-05" is located, and the directly connected environmental parameter feature nodes "daily average temperature" and "relative humidity" are found. The feature values ​​recorded in historical data for the environmental parameter feature node "daily average temperature" are queried. These feature values ​​include the historical maximum value "30℃", the historical minimum value "20℃", the historical average value "25℃", and the common value distribution range "23℃ to 27℃". The actual monitored value of the environmental parameter "26.5℃" was compared with the historical maximum value "30℃", the historical minimum value "20℃", and the historical average value "25℃" to calculate the absolute difference. The absolute difference with the historical maximum value of the environmental parameter characteristic node was "3.5℃", the absolute difference with the historical minimum value of the environmental parameter characteristic node was "6.5℃", and the absolute difference with the historical average value of the environmental parameter characteristic node was "1.5℃". Dividing the absolute difference "3.5℃" by the corresponding historical characteristic value change range "10℃" yields the relative deviation ratio "0.35" for the historical maximum value of the environmental parameter characteristic node. Dividing the absolute difference "6.5℃" by the historical characteristic value change range "10℃" yields the relative deviation ratio "0.65" for the historical minimum value of the environmental parameter characteristic node. Dividing the absolute difference "1.5℃" by the historical characteristic value change range "10℃" yields the relative deviation ratio "0.15" for the historical average value of the environmental parameter characteristic node. The maximum value of the relative deviation ratios "0.35", "0.65" and "0.15", "0.65", is selected as the final deviation of the environmental parameter characteristic node "daily average temperature" at the current moment.

[0112] In the dynamic coupling subgraph, all directed edges originating from the environmental parameter feature node "daily average temperature" and pointing to different historical feature nodes of pest occurrence are identified. For example, directed edges pointing to the historical feature node "aphids" and the historical feature node "powdery mildew" are found. The association strength weight attached to each directed edge is obtained. The association strength weight is calculated based on historical co-occurrence and transition frequency. The association strength weight of the directed edge pointing to the historical feature node "aphids" is "0.8", and the association strength weight of the directed edge pointing to the historical feature node "powdery mildew" is "0.3". The final deviation of the calculated environmental parameter feature node "daily average temperature" ("0.65") is used as an influencing factor. The historical occurrence frequency of each pest / disease occurrence historical feature node pointed to by each directed connection edge is used as another influencing factor. The historical occurrence frequency of the pest / disease occurrence historical feature node "aphid" is "0.4", and the historical occurrence frequency of the pest / disease occurrence historical feature node "powdery mildew" is "0.1". In specific implementation, the association strength weight, the final deviation influencing factor, and the historical occurrence frequency influencing factor are multiplied to obtain the comprehensive occurrence index for each possible pest / disease occurrence historical feature node. The calculation formula is as follows:

[0113]

[0114] in: It represents a comprehensive occurrence index for a specific historical characteristic node in the occurrence of a pest or disease. This represents the association strength weight attached to the directed connection edge pointing to this node, obtained from the dynamic coupling subgraph. This represents the final deviation of the calculated environmental parameter feature nodes. This represents the occurrence frequency of the historical characteristic node of the pest / disease occurrence, statistically obtained from historical data. For the historical characteristic node "aphids," its comprehensive occurrence index... For the historical characteristic node of the occurrence of pests and diseases, "powdery mildew", its comprehensive occurrence index All historical feature nodes of pest and disease occurrence are sorted from high to low according to their comprehensive occurrence index, resulting in the sorted results "Aphids (0.208)" and "Powdery Mildew (0.0195)". Based on a preset index threshold range, for example, a comprehensive occurrence index greater than 0.2 is mapped to "high probability", between 0.1 and 0.2 to "medium probability", and less than 0.1 to "low probability". The comprehensive occurrence index "0.208" is mapped to the level label "high probability", and the comprehensive occurrence index "0.0195" is mapped to the level label "low probability", thus generating the probability level of pest and disease occurrence. In some embodiments, the historical occurrence frequency is calculated based on the proportion of the number of times the pest or disease node appeared in all relevant growth stages in the past to the total number of observations.

[0115] See Figure 5 This is a bar chart comparing the frequency of pest and disease occurrence with the overall occurrence index. Aphids have the highest occurrence frequency (0.4) and overall occurrence index (0.21), making them the most prominent pest and disease currently monitored. Cabbage caterpillars have a relatively high occurrence frequency (0.25), but a relatively low overall occurrence index (0.13), indicating limited damage. Powdery mildew has a low occurrence frequency (0.1) and low overall occurrence index (0.02), classifying it as a low-risk pest and disease. Root rot has a higher occurrence frequency (0.1) than the overall occurrence index (0.05), reflecting its occurrence but relatively minor damage. This type of chart is mainly used for agricultural pest and disease risk assessment, quickly identifying high-risk pests and diseases (such as aphids) and providing a basis for priority control.

[0116] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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.

[0117] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A knowledge graph-based decision-making method for pest and disease control in the vegetable growth cycle, characterized in that, The method includes: Unstructured agricultural data streams containing vegetable varieties, growth stages, environmental parameters, pest and disease records, and geographic information were collected from the original database. The unstructured agricultural data stream is scanned line by line to identify the segments in the data stream that are associated with the vegetable growth cycle, and the content of each segment is extracted to generate a set of data tuples containing time stamps; The data tuple set is input into the feature fusion channel, and stage clustering is performed on the data tuples according to the vegetable growth stage to which the segment unit belongs, generating stage feature clusters sorted by growth cycle. Multi-dimensional feature association analysis was performed on the aforementioned stage feature clusters to establish a dynamic coupling relationship between environmental parameter features, pest and disease occurrence history features, and geospatial features under the growth cycle dimension. Based on the dynamic coupling relationship, decision parameters for pest and disease control of vegetables in a specific planting area at each stage of the growth cycle are generated.

2. The knowledge graph-based decision-making method for pest and disease control in the vegetable growth cycle according to claim 1, characterized in that, The unstructured agricultural data stream is scanned line by line to identify segments related to the vegetable growth cycle. Content is extracted from each segment to generate a set of data tuples containing time stamps, including: Receive real-time or historical unstructured agricultural data streams output from the raw database through a data interface; A sliding analysis window is set on the unstructured agricultural data stream, and keyword matching is performed on the data content passing through the sliding analysis window. The keywords include feature descriptors of vegetable growth cycle stages. When the data content in the sliding analysis window successfully matches any feature descriptor, the data content before and after the matching position with a preset length is extracted as a complete segment unit. Entity and attribute extraction is performed on each segment unit to identify the vegetable variety name, specific growth stage, pest and disease type, occurrence time and geographical location description text, and the extraction results are encapsulated into a structured data tuple. Each structured data tuple is added with a timestamp indicating its original appearance in the unstructured agricultural data stream, ultimately generating a set of data tuples containing the timestamps.

3. The knowledge graph-based decision-making method for pest and disease control in the vegetable growth cycle according to claim 2, characterized in that, The data tuple set is input into the feature fusion channel. Based on the vegetable growth stage to which the segment unit belongs, the data tuples are subjected to stage clustering to generate stage feature clusters sorted by growth cycle, including: Establish a feature fusion channel, which contains multiple parallel feature processing branches; Read the specific growth stage marked for each data tuple from the set of data tuples containing time stamps; Based on the specific growth stage, the data tuples are routed to the corresponding feature processing branches. Within each feature processing branch, data tuples belonging to the same specific growth stage are rearranged according to the order of their time stamps. For the rearranged data tuples, feature vectorization transformation is performed based on the pest and disease type and environmental parameter category represented internally, and vectorization results with similar features within the same time period are aggregated into a feature subset; The feature subsets output from all feature processing branches are assembled according to the natural order of their corresponding specific growth stages in the entire growth cycle of vegetables, generating stage feature clusters sorted by growth cycle.

4. The knowledge graph-based decision-making method for pest and disease control in the vegetable growth cycle according to claim 3, characterized in that, Multi-dimensional feature association analysis is performed on the aforementioned stage feature clusters to establish a dynamic coupling relationship between environmental parameter features, pest and disease occurrence history features, and geospatial features under the growth cycle dimension, including: The stage feature clusters ordered by the growth cycle are analyzed to separate environmental parameter feature subsequences, pest and disease occurrence history feature subsequences, and geospatial feature subsequences. For each specific growth stage, create a dynamic association graph model corresponding to that growth stage; In the dynamic association graph model, the feature values ​​of the environmental parameter feature subsequence in the growth stage are used as environmental nodes, the feature values ​​of the pest and disease occurrence history feature subsequence in the growth stage are used as pest and disease nodes, and the feature values ​​of the geospatial feature subsequence in the growth stage are used as spatial nodes. By analyzing the co-occurrence and transfer patterns among environmental nodes, pest and disease nodes, and spatial nodes in historical data, directed connection edges among these nodes are defined and created. Each of the directed connecting edges is accompanied by an association strength weight calculated from historical statistical frequencies, thereby forming a dynamic coupled subgraph describing the multidimensional feature interaction relationships within the growth stage; The dynamic coupling subgraphs corresponding to each growth stage are merged according to the growth cycle order to generate dynamic coupling relationships that run through the entire growth cycle. The analysis of the stage feature clusters ordered by growth cycle is used to separate environmental parameter feature subsequences, pest and disease occurrence history feature subsequences, and geospatial feature subsequences, including: Load the stage feature clusters sorted by growth cycle, and deconstruct the feature subsets corresponding to each growth stage in the stage feature clusters; From each feature subset, data labeled with feature vectors such as temperature, humidity, light intensity, and soil pH are identified and extracted, and arranged in chronological order to form the environmental parameter feature fragments of the growth stage. From the same feature subset, identify and extract data whose feature vector labels are the name of the pest or disease, the time of occurrence, and the severity, and arrange them in chronological order to form a historical feature segment of the occurrence of pests and diseases in the growth stage. From the same feature subset, identify and extract data whose feature vector labels are latitude and longitude coordinates, altitude, and plot number, and arrange them in chronological order to form the geospatial feature fragments of the growth stage. The environmental parameter feature segments, pest and disease occurrence history feature segments, and geospatial feature segments corresponding to each growth stage are vertically concatenated according to their growth stage order to generate environmental parameter feature subsequences, pest and disease occurrence history feature subsequences, and geospatial feature subsequences.

5. The knowledge graph-based decision-making method for pest and disease control in the vegetable growth cycle according to claim 4, characterized in that, By analyzing the co-occurrence and transfer patterns among environmental nodes, pest and disease nodes, and spatial nodes in historical data, directed connections between these nodes are defined and created, including: Collect records of the environmental parameter feature subsequence, the pest and disease occurrence history feature subsequence, and the geospatial feature subsequence in multiple historical growth cycles; The co-occurrence frequency is calculated as the number of times that a specific environmental parameter characteristic node, a specific pest and disease occurrence history characteristic node, and a specific geospatial characteristic node occur simultaneously within the same growth stage. The frequency of transitions is calculated as the number of times a specific environmental parameter characteristic node transitions to a specific historical characteristic node of a specific disease or pest occurrence between adjacent growth stages, and the number of times a specific geographic spatial characteristic node transitions to a specific historical characteristic node of a specific disease or pest occurrence. Set a co-occurrence frequency threshold and a transfer frequency threshold. When the co-occurrence frequency exceeds the threshold, create undirected connection edges between each pair of the environmental node, pest node, and spatial node to indicate the existence of an association. When the frequency of the transition exceeds its threshold, a directed connection edge representing a causal or influence relationship is created from the node in the starting state to the node in the arriving state. The directed and undirected edges are merged to form the set of connecting edges between nodes in the growth stage.

6. The knowledge graph-based decision-making method for pest and disease control in the vegetable growth cycle according to claim 5, characterized in that, The dynamic coupling subgraphs corresponding to each growth stage are fused according to the growth cycle sequence to generate dynamic coupling relationships that span the entire growth cycle, including: Obtain the dynamic coupling subgraph corresponding to the first growth stage and use it as the initial fusion graph; Read the dynamic coupling subgraph corresponding to the next growth stage in the order of the growth cycle. Identify the same nodes in the initial fusion graph and the dynamic coupling subgraph of the next growth stage, where the same nodes refer to the same geospatial features or the same vegetable varieties; Merge the nodes that are the same as those in the initial fusion graph in the next growth stage dynamic coupling subgraph. In the next growth stage dynamic coupling subgraph, unique new nodes and new connecting edges are added to the initial fusion graph; In the merged graph, between nodes of adjacent growth stages, temporal connection edges representing the cross-stage evolution of pests and diseases are added based on the sub-sequence of the historical characteristics of the occurrence of pests and diseases. Repeat the process until all the dynamic coupling subgraphs of all growth stages are merged, ultimately generating a dynamic coupling relationship graph that integrates all feature nodes and complex connections throughout the entire growth cycle.

7. The knowledge graph-based decision-making method for pest and disease control in the vegetable growth cycle according to claim 4, characterized in that, Based on the aforementioned dynamic coupling relationship, decision parameters for pest and disease control of vegetables in a specific planting area at each stage of the growth cycle are generated, including: Receive a query request for a target vegetable variety in a specific growing area, the query request containing a geolocation code; Traverse the dynamic coupling relationship, filter out the geospatial feature nodes that match the geographic location code, and locate the growth stage to which the geospatial feature node belongs; Extract the dynamic coupling subgraph corresponding to the growth stage, and obtain all environmental parameter feature nodes and pest and disease occurrence history feature nodes that have directed connection edges with the geospatial feature nodes; Read the current monitoring value of the environmental parameter feature node, compare it with the historical feature value range of the environmental parameter feature node recorded in the dynamic coupling subgraph, and calculate the deviation. By combining the association strength weights on the directed connection edges, the historical feature nodes of pest and disease occurrence associated with the current environmental monitoring value are weighted and evaluated to predict the probability level of pest and disease occurrence. Based on the probability level of the occurrence of pests and diseases and the associated pest and disease types, pest and disease control decision parameters including the timing of prevention and control, recommended measures and intensity levels are generated from a preset strategy library.

8. The knowledge graph-based decision-making method for pest and disease control in the vegetable growth cycle according to claim 7, characterized in that, Read the current monitoring value of the environmental parameter feature node, compare it with the historical feature value range of the environmental parameter feature node recorded in the dynamic coupling subgraph, and calculate the deviation, including: Obtain the actual monitored values ​​of environmental parameters for the specific planting area at the current growth stage from a real-time data source; From the dynamic coupling subgraph, locate the geospatial feature node corresponding to the geographic location code, and find the environmental parameter feature node directly connected to it; Query the feature values ​​of the environmental parameter feature nodes recorded in historical data. The feature values ​​include the maximum value, minimum value, average value, and common numerical distribution ranges. The actual monitored values ​​of the environmental parameters are compared with the historical maximum, historical minimum, and historical average values, and the absolute difference is calculated. Divide the absolute difference by the corresponding historical feature value change range to obtain multiple relative deviation ratios for the maximum, minimum, and average values; The maximum value among the multiple relative deviation ratios is selected as the final deviation of the environmental parameter feature node at the current moment.

9. The knowledge graph-based decision-making method for pest and disease control in the vegetable growth cycle according to claim 8, characterized in that, By combining the association strength weights on the directed connection edges, a weighted evaluation is performed on the historical characteristic nodes of pest and disease occurrence associated with the current environmental monitoring values ​​to predict the probability level of pest and disease occurrence, including: In the dynamic coupling subgraph, find all directed connections that start from the environmental parameter feature nodes and point to different pest and disease occurrence history feature nodes. Obtain the association strength weight attached to each of the directed connection edges, the association strength weight being calculated based on historical co-occurrence and transition frequency; The final deviation of the calculated environmental parameter feature nodes is used as an influencing factor; The historical occurrence frequency of each disease and pest occurrence node pointed to by each directed connection edge is used as another influencing factor. Multiply the correlation strength weight, the final deviation influence factor, and the historical occurrence frequency influence factor to obtain a comprehensive occurrence index for each possible historical characteristic node of pest and disease occurrence; Sort all historical characteristic nodes of pest and disease occurrence from high to low according to the comprehensive occurrence index; Based on a preset index threshold range, the comprehensive occurrence index is mapped to different level labels, including high probability, medium probability, and low probability, thereby generating a pest and disease occurrence probability level.

10. A knowledge graph-based decision-making system for the prevention and control of diseases and pests in the vegetable growth cycle, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the knowledge graph-based decision-making method for the prevention and control of diseases and pests in the vegetable growth cycle as described in any one of claims 1 to 9.

Images