Compatibility relation real-time query method and system based on pesticide knowledge graph

Abstract

The invention discloses a pesticide knowledge graph-based compatibility relation real-time query method and system, and relates to the technical field of knowledge graphs, and the method comprises the steps: recognizing a pesticide entity, and constructing a pesticide knowledge graph based on dynamics priori knowledge; constructing and training a compatibility relation mapping model; and carrying out relation evaluation in combination with a pesticide knowledge map and the compatibility relation mapping model to obtain a compatibility relation vector, and correspondingly outputting a compatibility query report. The technical problems that an existing pesticide compatibility query method is insufficient in accuracy, slow in updating, tedious in query and low in efficiency are solved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of knowledge graphs, in particular to a compatibility relationship real-time query method and system based on a pesticide knowledge graph. BACKGROUND

[0002] With the acceleration of agricultural modernization, pesticides play a crucial role in ensuring crop yields and preventing pests and diseases. In order to improve the control effect, expand the control spectrum or reduce the use cost, different types of pesticides are often mixed and used in agricultural production. However, there are many types of pesticides, their effective components and physicochemical properties are different, and the compatibility relationship between different pesticides is complex and variable, which may produce synergistic effect, additive effect or antagonistic effect, and even cause safety problems such as pesticide damage and excessive residue.

[0003] Traditional pesticide compatibility query methods mostly rely on the experience accumulation of agricultural technicians or the consultation of static data such as paper manuals and databases, which is difficult to ensure accuracy. Moreover, static data is updated slowly and cannot reflect the latest research results and new pesticide products emerging on the market in real time. In addition, the query process is tedious and inefficient, which cannot meet the demand for fast and accurate acquisition of pesticide compatibility information in modern agricultural production. SUMMARY

[0004] The embodiment of the application provides a compatibility relationship real-time query method and system based on a pesticide knowledge graph, which solves the technical problems of the prior art, such as insufficient accuracy, slow updating, tedious query and low efficiency.

[0005] The technical solution of the application to solve the above technical problems is as follows: In a first aspect, the application provides a compatibility relationship real-time query method based on a pesticide knowledge graph, which comprises: identifying pesticide entities and constructing a pesticide knowledge graph based on kinetic prior knowledge; constructing and training a compatibility relationship mapping model based on historical compatibility experiment data and the pesticide knowledge graph; in response to a user inputted pesticide combination, performing relationship evaluation in combination with the pesticide knowledge graph and the compatibility relationship mapping model, obtaining a compatibility relationship vector, and outputting a compatibility query report correspondingly.

[0006] In a second aspect, the application provides a compatibility relationship real-time query system based on a pesticide knowledge graph, which comprises: a graph identification module configured to identify pesticide entities and construct a pesticide knowledge graph based on kinetic prior knowledge; a model construction module configured to construct and train a compatibility relationship mapping model based on historical compatibility experiment data and the pesticide knowledge graph; The report output module is configured to, in response to a user inputted pesticide combination to be tested, perform relationship evaluation in combination with the pesticide knowledge graph and the compatibility relationship mapping model, obtain a compatibility relationship vector, and correspondingly output a compatibility query report.

[0007] The present application provides one or more technical solutions, at least having the following technical effects or advantages: The present application provides a compatibility relationship real-time query method and system based on a pesticide knowledge graph. Firstly, pesticide entities are identified and a pesticide knowledge graph based on kinetic prior knowledge is constructed to provide data support for subsequent compatibility relationship evaluation. Secondly, based on historical compatibility experiment data and the pesticide knowledge graph constructed above, a compatibility relationship mapping model is constructed and trained. The original experiment data is filtered and cleaned by calculating multi-dimensional indexes such as scene similarity, zoning similarity and time distance, and the model capable of mapping the kinetic characteristics between pesticide entities and the compatibility relationship is obtained by machine learning. Finally, when responding to a user inputted pesticide combination to be tested, the target kinetic characteristic parameters of each entity in the tested combination are obtained in combination with the pesticide knowledge graph and inputted into the trained compatibility relationship mapping model for relationship evaluation, a quantitative compatibility relationship vector is generated, and a compatibility query report containing visual results and natural language explanation is outputted.

[0008] Through the above technical solutions, the present application effectively improves the accuracy and real-time performance of pesticide compatibility query based on the structured storage and dynamic updating characteristics of the knowledge graph and the intelligent reasoning ability of the mapping model, improves the query efficiency, and thus better meets the demand for fast and accurate acquisition of pesticide compatibility information in modern agricultural production. BRIEF DESCRIPTION OF DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0010] Figure 1 is a flowchart of the compatibility relationship real-time query method based on a pesticide knowledge graph provided by the embodiments of the present application; Figure 2 is a structural diagram of the compatibility relationship real-time query system based on a pesticide knowledge graph provided by the embodiments of the present application.

[0011] In the drawings, the components represented by the numbers are described as follows: The graph identification module 11, the model construction module 12, and the report output module 13. DETAILED DESCRIPTION

[0012] The embodiment of the application provides a compatibility relationship real-time query method and system based on a pesticide knowledge graph, and is used for solving the technical problems of low accuracy, slow updating, complicated query and low efficiency of an existing pesticide compatibility query method.

[0013] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person skilled in the art without creative work fall within the protection scope of the application.

[0014] In the description of the application, the terms "first", "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first", "second" can explicitly or implicitly include one or more of the features. In the description of the application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0015] In the description of the application, the term "for example" is used to indicate "as an example, illustration or explanation". Any embodiment described as "for example" in the application is not necessarily interpreted as more preferred or more advantageous than other embodiments. In order to enable any person skilled in the art to implement and use the application, the following description is given. In the following description, details are listed for the purpose of explanation. It should be understood that a person skilled in the art can realize the application without using these specific details. In other examples, well-known structures and processes will not be described in detail to avoid unnecessary details making the description of the application obscure. Therefore, the application is not intended to be limited to the shown embodiments, but is consistent with the broadest scope of the principles and characteristics disclosed in the application.

[0016] Embodiment one, as Figure 1 shown, the embodiment of the application provides a compatibility relationship real-time query method based on a pesticide knowledge graph, comprising: S10: identifying pesticide entities and constructing a pesticide knowledge graph based on kinetic prior knowledge; Wherein, identifying pesticide entities comprises: determining a plurality of pesticide products under a target scene, and corresponding extracting ingredient information of the plurality of pesticide products; summarizing the ingredient information and extracting a plurality of main ingredients with pharmacological effects, defined as a plurality of absolute pesticide entities; traversing a plurality of the absolute pesticide entities, identifying a plurality of adjuvant ingredients with physicochemical reaction characteristics existing in the absolute pesticide entities, defined as a plurality of relative pesticide entities; The combined output of multiple relative pesticide entities and multiple absolute pesticide entities is the pesticide entity identification result.

[0017] In this embodiment of the application, firstly, pesticide entities are identified and a pesticide knowledge graph based on kinetic prior knowledge is constructed. Several pesticide products actually used in a certain scenario are first determined. For example, for the scenario of pest and disease control in rice fields, commonly used pesticide products such as insecticides and fungicides on the market are collected.

[0018] Then, the ingredient information of pesticide products is prepared in advance, including all components such as active ingredients and adjuvants listed in the product instructions.

[0019] Subsequently, all extracted component information was summarized and organized. Using pesticide science knowledge and data analysis methods, several main components with clear pharmacological efficacy and core roles in pest and disease control or crop growth regulation were screened out, and the main components were defined as absolute pesticide entities.

[0020] Secondly, based on the absolute pesticide entity, a further comprehensive analysis was conducted to identify multiple adjuvant components in pesticide products that have specific physicochemical reaction characteristics with the absolute pesticide entity. For example, adjuvants that can promote the absorption of the absolute pesticide entity, enhance its stability, or change its physical state are defined as relative pesticide entities.

[0021] Finally, the identified relative pesticide entities and absolute pesticide entities are merged to form a complete pesticide entity identification result, providing an accurate entity foundation for the subsequent construction of a pesticide knowledge graph.

[0022] Furthermore, a pesticide knowledge graph based on kinetic prior knowledge is constructed, including: Using the identified pesticide entity as an index, the kinetic prior knowledge related to the pesticide entity is retrieved and obtained from a predefined multi-source data source, wherein the kinetic prior knowledge includes unstructured text data and structured data; The unstructured text data and structured data are input into a pre-trained parameter parsing-normalization engine to obtain quantitative parameters of multiple pesticide entities in a preset compatibility relationship dimension. Based on the quantization parameters, a corresponding multidimensional array is generated for each pesticide entity, wherein the multidimensional array contains scalar values; Using the pesticide entity as a node and the multidimensional array as the relationship between multiple nodes, the pesticide knowledge graph is obtained by connecting multiple nodes and storing them in a structured manner.

[0023] In this embodiment, the identified pesticide entity is first used as the core index to retrieve kinetic prior knowledge related to the pesticide entity from pre-defined multi-source data sources. These multi-source data sources include pesticide databases, published academic research literature, detailed technical manuals for pesticide products, and verified experimental reports.

[0024] Among them, kinetic prior knowledge includes unstructured textual data, such as descriptive texts on the mechanism of action of pesticides in academic papers, records of experimental phenomena, etc., as well as structured data, such as numerical data that can be directly quantified, such as the reaction rate constant, half-life, and solubility of pesticides under different conditions.

[0025] Secondly, the collected unstructured text data and structured data are input into a pre-trained parameter parsing and standardization engine. This engine integrates natural language processing technology and data standardization algorithms, enabling deep semantic understanding and key information extraction from unstructured text, transforming it into structured parameters. It establishes clear quantitative standards and dimensions for each dimension, providing a basis for subsequent numerical filling. Simultaneously, it performs standardization operations on the existing structured data, including format unification, unit conversion, and outlier handling, ultimately obtaining quantitative parameters for multiple pesticide entities across preset compatibility dimensions. Compatibility dimensions may include, but are not limited to, chemical stability, synergistic / antagonistic bioactivity, physical compatibility, and toxicity trends, with each dimension corresponding to one or more specific quantitative parameters.

[0026] Then, based on the parsed and standardized quantization parameters, a corresponding multidimensional array is generated for each pesticide entity. Each element in this multidimensional array is a scalar value, corresponding to the quantization parameter value of the pesticide entity in a specific compatibility dimension. For example, a multidimensional array may contain scalar information such as the activation energy of the reaction when a pesticide entity is mixed with other entities, the maximum synergistic effect percentage, and the change in suspension rate.

[0027] Finally, the identified pesticide entities are used as nodes in the knowledge graph, and the multidimensional array corresponding to each pesticide entity is used as the description and quantification basis for the relationship between nodes, thus connecting multiple nodes.

[0028] Specifically, when a potential compatibility relationship exists between two pesticide entities, the dimension parameters related to this compatibility in the multidimensional array will collectively constitute the attributes describing the relationship between the two. The nodes and their relationships are then stored in a structured manner according to a graph data structure, such as using a graph database like Neo4j, thereby constructing a pesticide knowledge graph based on kinetic prior knowledge.

[0029] S20: Based on historical compatibility experimental data and the pesticide knowledge graph, construct and train a compatibility relationship mapping model; Among these steps, based on historical compatibility experimental data and the aforementioned pesticide knowledge graph, a compatibility relationship mapping model is constructed and trained. Prior to this, the following steps were taken: Obtain the original compatibility experiment dataset containing multiple original compatibility experiment items, wherein each original compatibility experiment item includes experimental crop information, experimental geographical division information, experimental time information, and experimental compatibility results; Calculate the scene similarity scalar of each original matching experiment item and the target scene in terms of crop variety, the regional similarity scalar in terms of geographical region, and the time distance scalar from the current time point; Based on preset weights, the scene similarity scalar, the region similarity scalar, and the time distance scalar are weighted and calculated to obtain the rejection probability value of each original matching experiment item. Based on the rejection probability value, the selected original matching experiment items are iteratively filtered and removed from the original matching experiment dataset; When the number of remaining original matching experiment items in the original matching experiment dataset after filtering reaches a preset threshold, the data cleaning process is terminated, and the original matching experiment dataset after filtering is output as the historical matching experiment data.

[0030] In this embodiment of the application, before constructing and training the compatibility mapping model based on historical compatibility experimental data and the pesticide knowledge graph, the first step is to obtain an original compatibility experimental dataset containing multiple original compatibility experimental items. The original data comes from pesticide field trial reports, indoor toxicity test data, and publicly released compatibility research results from agricultural research institutions. Each original compatibility experimental item records experimental crop information, such as crop type, variety name, and growth stage; experimental geographical division information, including specific administrative divisions, latitude and longitude ranges, climate zone type, and soil type; experimental time information, i.e., the specific year and season in which the experiment was conducted; and experimental compatibility results, which typically include the control effect of the pesticide combination on target pests and diseases, the safety evaluation of crops, and possible physicochemical reaction phenomena, such as whether stratification, precipitation, or heating occurs.

[0031] Secondly, the scene similarity scalar between each original pesticide compatibility experiment item and the target scene in terms of crop varieties is calculated. The target scene is the specific application scenario involved when the user is currently performing a pesticide compatibility query. The scene similarity can be calculated by constructing a crop variety feature vector, which includes attributes such as crop classification, morphological characteristics, physiological characteristics, and common pest and disease types. The cosine similarity algorithm is used to calculate the similarity between the original experimental item crop variety feature vector and the target scene crop variety feature vector to obtain the scene similarity scalar, whose value ranges from [0,1]. The closer the value is to 1, the higher the similarity.

[0032] Simultaneously, the scalar similarity between the original matching experimental items in terms of geographical regions is calculated. First, the geographical regions of both the experimental and target scenarios are converted into vectors containing multi-dimensional features such as climate factors, soil physicochemical properties including pH value, organic matter content, nitrogen, phosphorus and potassium content, and ecological region division. Then, the cosine similarity algorithm or the weighted similarity algorithm based on the analytic hierarchy process is used to calculate the scalar similarity between the two, with a value range of [0,1].

[0033] Next, calculate the time distance scalar between the original matched experimental items and the current time point. The time distance scalar can be defined as the reciprocal of the difference between the current year and the year the experiment was conducted. For example, if the current year is 2024 and an experiment was conducted in 2022, then the time distance scalar is 1 / (2024-2022)=0.5. The closer the time, the larger the scalar value and the higher the data weight.

[0034] Then, the scene similarity scalar, region similarity scalar, and time distance scalar are weighted according to preset weights to obtain the rejection probability value for each original matching experiment item. The preset weights can be adjusted according to the importance of the target scene. In scenarios where crop varieties are sensitive to pesticides, the weight of scene similarity can be set to 0.5, region similarity to 0.3, and time distance to 0.2. The formula for calculating the rejection probability value can be set as: rejection probability value = 1 - (scene similarity scalar × scene weight + region similarity scalar × region weight + time distance scalar × time weight).

[0035] For example, if the scene similarity scalar of an original matching experiment item is 0.8, the region similarity scalar is 0.6, the time distance scalar is 0.7, the scene weight is 0.5, the region weight is 0.3, and the time weight is 0.2, then its rejection probability value = 1 - (0.8 × 0.5 + 0.6 × 0.3 + 0.7 × 0.2) = 0.28. The higher the rejection probability value, the weaker the correlation between the original experiment item and the target scene, and the more likely it is to be eliminated.

[0036] Furthermore, based on the calculated rejection probability values, the original matching experiment dataset is iteratively filtered according to probability magnitude. Each time, a subset of original matching experiment items with higher rejection probabilities are randomly selected and removed. During the iterative filtering process, the number of remaining original matching experiment items is continuously monitored. When the remaining number reaches a preset threshold (e.g., retaining 60%-80% of the total original data as valid data), the data cleaning process is terminated. The remaining original matching experiment items in the filtered dataset then constitute the historical matching experiment data used for model training. This ensures that the training data input to the model has high scenario adaptability and timeliness, improving the model's prediction accuracy for real-world application scenarios.

[0037] Specifically, step S20 in the method includes: Combining the preset compatibility relationship dimensions, the historical compatibility experimental data is analyzed to evaluate the compatibility relationship and obtain the sample compatibility relationship vector; Based on the historical compatibility experiment data, extract the sample pesticide entity set from each historical compatibility experiment item; Using the pesticide entity set as an index, the pesticide knowledge graph is traversed for entity matching, and a sample dynamic feature parameter set for each pesticide entity set is defined based on the entity matching results. By combining the sample compatibility vector with the sample dynamics feature parameter set, the compatibility mapping model is constructed and trained based on machine learning methods.

[0038] The sample matching relationship vector is associated with at least one of the sample multidimensional dynamic feature vectors.

[0039] In this embodiment, firstly, historical compatibility experimental data are deeply analyzed to complete the compatibility evaluation by combining preset compatibility relationship dimensions, such as predefined discrete relationships like chemical compatibility, antagonism, synergy, ecological risk, and resistance risk. The preset compatibility relationship dimensions are consistent with the dimensions used when constructing the pesticide knowledge graph, such as chemical stability, degree of synergy / antagonism of biological activity, physical compatibility, and toxicity change trends. For each historical compatibility experiment, based on its experimental results, such as control effect data, crop phytotoxicity, records of changes in pesticide appearance, and toxicity test values, each compatibility relationship dimension is quantitatively scored or graded.

[0040] For example, in the dimension of synergistic bioactivity, if the experimental results show that the control effect of the mixture of two pesticides is 30% higher than that of the mixture alone, the evaluation result of this dimension can be quantified as 0.8, with a maximum score of 1.0; if there is significant antagonism and the control effect decreases by 20%, it can be quantified as -0.3. The evaluation results of all dimensions are integrated to form a sample compatibility vector that reflects the compatibility relationship between pesticide entities in this historical compatibility experiment.

[0041] Secondly, based on the organized historical compatibility test data, the pesticide products involved in each historical compatibility test item are extracted, and specific pesticide entities are extracted from the pesticide products according to the established pesticide entity identification rules to form a sample pesticide entity set.

[0042] Then, using the extracted sample pesticide entity set as a retrieval index, the constructed pesticide knowledge graph is traversed for entity matching. If a corresponding pesticide entity node is matched in the knowledge graph, all relevant kinetic feature parameter values ​​are extracted from the multidimensional array associated with that node, i.e., extracted from the kinetic feature parameter set. The kinetic feature parameter sets of all entities in the sample pesticide entity set are summarized and organized to obtain the sample kinetic feature parameter set corresponding to that sample pesticide entity set.

[0043] Finally, the sample compatibility vector is used as the target output label for model training, and the corresponding set of sample kinetic feature parameters is used as the input features of the model. A machine learning method is then used to construct and train the compatibility mapping model. A deep neural network machine learning algorithm is chosen for model construction. The model's input layer receives the set of sample kinetic feature parameters, which contains the multidimensional kinetic features of multiple pesticide entities. The model's output layer outputs a prediction vector corresponding to the sample compatibility vector structure. During training, by adjusting the model parameters and minimizing the loss function between the prediction vector and the sample compatibility vector, the model learns the complex mapping relationship between the kinetic feature parameters of pesticide entities and their compatibility evaluation results.

[0044] Furthermore, a sample compatibility vector may correspond to at least one sample multidimensional kinetic feature vector, indicating that the model needs to be able to handle various combinations of pesticide entity inputs and predict their comprehensive compatibility relationships. For example, a sample may involve the compatibility of two pesticide entities. In this case, the sample kinetic feature parameter set will contain the multidimensional kinetic feature vectors of each of the two entities, and the model needs to learn by using both vectors as a whole input. Through training with a large number of such samples, the compatibility mapping model gradually acquires the ability to predict the compatibility relationships of unknown pesticide entity combinations in various dimensions based on their kinetic feature parameters.

[0045] Furthermore, combining the sample compatibility vector with the sample dynamics feature parameter set, a compatibility mapping model is constructed and trained based on machine learning methods, including: Vectorize the sample dynamic feature parameter set to obtain the sample multidimensional dynamic feature vector; Establish the correspondence between the sample matching relationship vector and the sample multidimensional dynamic feature vector; Initialize the matching relationship mapping model based on machine learning, and according to the correspondence, use the multidimensional dynamic feature vector of the sample as the sample input and the matching relationship vector of the sample as the sample output to perform supervised training on the matching relationship mapping model until the preset termination constraint is met.

[0046] In this embodiment, the sample dynamics feature parameter set is first vectorized to convert it into a numerical vector form suitable for input to a machine learning model, i.e., a sample multidimensional dynamics feature vector. The sample dynamics feature parameter set typically contains multidimensional array information for multiple pesticide entities. For example, when the sample pesticide entity set contains two pesticide entities, each entity corresponds to a multidimensional array. The two multidimensional arrays are concatenated sequentially to form a longer one-dimensional vector, which yields the sample multidimensional dynamics feature vector corresponding to that sample. If a sample pesticide entity set contains n pesticide entities, and the length of the multidimensional array for each entity is m, then the total length of the sample multidimensional dynamics feature vector is n×m.

[0047] For example, if the multidimensional array of entity A is [0.5, 0.3, -0.2] and the multidimensional array of entity B is [0.7, -0.1, 0.4], then the concatenated sample multidimensional dynamic feature vector is [0.5, 0.3, -0.2, 0.7, -0.1, 0.4]. By vectorizing, the dynamic feature parameters of multiple entities are integrated into a single vector representation, which facilitates unified feature learning and processing by the model.

[0048] Secondly, a one-to-one correspondence is established between the sample compatibility relationship vector and the sample multidimensional dynamic feature vector. Each historical compatibility experiment item corresponds to a set of sample pesticide entities. After entity matching and feature extraction, the sample multidimensional dynamic feature vector is obtained. At the same time, the sample compatibility relationship vector is obtained based on the analysis of the experimental results. The two vectors together constitute a complete training sample pair.

[0049] For example, in a certain historical experiment, the multidimensional dynamic feature vector of pesticide entities A and B is V, and the corresponding sample matching relationship vector is R. Then, a mapping relationship of V→R is established so that the model can learn how to output a prediction vector that is close to R when the input is V.

[0050] Next, initialize the machine learning-based matchmaking mapping model. Select a suitable model structure, such as a deep learning model, based on task requirements and data characteristics. Initialize the weight parameters and biases of each layer, and set hyperparameters such as the number of network layers, the number of neurons in each layer, and the activation function type. For example, choose a network structure containing an input layer, two hidden layers, and an output layer. The number of nodes in the input layer should match the length of the sample's multidimensional dynamic feature vector, and the number of nodes in the output layer should match the dimension of the sample's matchmaking vector. Set the hidden layers to 128 and 64 neurons respectively, and use ReLU as the activation function.

[0051] Finally, based on the established correspondence between the sample matching relationship vector and the sample multidimensional dynamic feature vector, the matching relationship mapping model is trained under supervision, using the sample multidimensional dynamic feature vector as input and the sample matching relationship vector as the desired output. During training, the training sample set is divided into a training set and a validation set. The training set is used to update the model parameters, the gradient of the loss function with respect to each parameter is calculated using the backpropagation algorithm, and an optimizer, such as Adam, is used to adjust the parameters to minimize the loss. Simultaneously, the validation set is used to monitor the model's generalization ability in real time. When the validation set loss meets preset termination constraints, such as the validation set loss no longer decreasing for several consecutive training epochs, reaching the preset maximum number of training epochs, or the loss value falling below a set threshold, model training is stopped.

[0052] Specifically, when constructing and training the compatibility mapping model, the number of nodes in the input layer equals the dimension of the input features. For example, if the sample dynamics feature parameter set has 4 features, the input layer contains 4 nodes. One to three hidden layers are set, with the number of nodes in each layer adjusted experimentally (e.g., 64, 32). The ReLU activation function is used. The output layer generally does not use an activation function; if the output takes 2 nodes, continuous values ​​are directly output. The Adam optimizer and mean squared error loss function are used to construct the training framework. The batch size is set to 32, the total training epochs to 50, and an early stopping mechanism (patience=5) is introduced. When the validation set loss does not decrease for 5 consecutive epochs, the training process is automatically terminated, resulting in the trained compatibility mapping model.

[0053] Through supervised training, the model gradually learns the inherent laws between the kinetic characteristic parameters and compatibility relationships of pesticide entities, thus enabling it to accurately predict the compatibility vector when it receives the kinetic characteristic parameters of a new pesticide entity combination.

[0054] S30: In response to the user input of the pesticide combination to be tested, the relationship is evaluated by combining the pesticide knowledge graph with the compatibility relationship mapping model, the compatibility relationship vector is obtained, and a compatibility query report is output accordingly.

[0055] In this embodiment, firstly, the pesticide combination to be tested is evaluated by combining a pesticide knowledge graph with a compatibility mapping model. This pesticide combination contains at least two pesticide products with compatibility relationships to be evaluated. Based on pesticide entity recognition rules, specific pesticide entities are extracted from the pesticide products to form a pesticide entity set to be tested.

[0056] Secondly, using the set of pesticide entities to be tested as an index, the constructed pesticide knowledge graph is traversed for entity matching. If a matching pesticide entity node exists in the knowledge graph, all relevant kinetic feature parameter values ​​are extracted from the multidimensional array associated with that node to form a set of kinetic feature parameters to be tested. The set of kinetic feature parameters to be tested is then vectorized, and the multidimensional arrays of multiple pesticide entities are sequentially concatenated into a single multidimensional kinetic feature vector to be tested.

[0057] Furthermore, the multidimensional dynamic feature vector to be tested is input into the trained compatibility mapping model. Through forward propagation calculation, the predicted vector corresponding to the preset compatibility dimension is output, which is the compatibility vector of the pesticide combination to be tested. Based on the output compatibility vector, the corresponding compatibility query report is generated.

[0058] Specifically, step S30 in the method includes: Obtain the pesticide combination to be tested input by the user, and extract the corresponding pesticide entity set; Based on the pesticide knowledge graph, the target kinetic feature parameter set corresponding to the pesticide entity set to be tested is obtained by matching, wherein the target kinetic feature parameter set includes target kinetic feature parameter clusters of multiple uniquely associated pesticide entity pairs; Vectorize multiple target dynamic feature parameter clusters to obtain multiple target multidimensional dynamic feature vectors, and then concatenate the vectors. The vector concatenation result is input into the matching relationship mapping model to obtain the output matching relationship vector.

[0059] In this embodiment, firstly, the system obtains the pesticide combination to be tested input by the user. The user can input information such as the name of the pesticide product to be queried, the active ingredient, or the registration certificate number through an interactive interface. Subsequently, based on the pesticide entity recognition rules mentioned above, specific pesticide entities are extracted from the pesticide products included in the pesticide combination to be tested input by the user. For example, the pesticide entity "imidacloprid" is extracted from "25% imidacloprid wettable powder," thus forming a set of pesticide entities to be tested. If the pesticide product input by the user cannot be recognized or a valid pesticide entity cannot be extracted in the system, a prompt message is returned, guiding the user to re-enter or check the input content.

[0060] Secondly, based on the constructed pesticide knowledge graph, entity matching is performed on the extracted pesticide entity set to obtain the corresponding target kinetic feature parameter set.

[0061] Specifically, each pesticide entity in the pesticide entity set to be tested is used as a search keyword, and the entity nodes in the pesticide knowledge graph are traversed. When a pesticide entity node is successfully matched in the knowledge graph, all parameter clusters related to kinetic features are extracted from the attribute information associated with that node. The target kinetic feature parameter clusters are consistent in dimension and type with the sample kinetic feature parameter set used when constructing the model. For each entity in the pesticide entity set to be tested, the above matching and extraction operations are performed, and the target kinetic feature parameter clusters corresponding to all entities are summarized to obtain the target kinetic feature parameter set corresponding to that pesticide entity set.

[0062] Then, the acquired multiple target kinetic feature parameter clusters are vectorized and concatenated. Similar to the method used when processing sample data, for each target kinetic feature parameter cluster, the values ​​of each kinetic feature parameter contained therein are arranged in a preset order and converted into a one-dimensional numerical vector, thus obtaining the target multidimensional kinetic feature vector corresponding to each pesticide entity.

[0063] For example, if a pesticide entity has a target kinetic characteristic parameter cluster including a molecular weight of 250, a solubility of 10 mg / L, an octanol-water partition coefficient of 3.5, and a degradation half-life of 15 days, then its corresponding target multidimensional kinetic characteristic vector can be represented as [250, 10, 3.5, 15].

[0064] Then, the target multidimensional dynamic feature vectors of all pesticide entities in the pesticide entity set to be tested are concatenated in the order of the entities in the set to form a one-dimensional vector, which is used as the input feature vector for subsequent model prediction.

[0065] For example, the pesticide entity set to be tested contains entity A and entity B, whose target multidimensional dynamic feature vectors are [A1,A2,A3] and [B1,B2,B3,B4], respectively. Then the concatenated vector is [A1,A2,A3,B1,B2,B3,B4].

[0066] Finally, the above vector concatenation result is used as input and fed into the trained compatibility mapping model. The model performs forward propagation calculations on the input concatenated vector, and through the operations of each layer of the neural network, it finally outputs a predicted vector corresponding to the preset compatibility dimension. This predicted vector is the compatibility vector of the pesticide combination to be tested. Each element value in this compatibility vector corresponds to the quantitative evaluation result of each preset compatibility dimension. For example, the score for the chemical compatibility dimension is 0.7, the score for the synergistic effect dimension is 0.6, and the score for the ecological risk dimension is 0.2. These values ​​comprehensively reflect the compatibility characteristics between the pesticide combinations to be tested.

[0067] Specifically, in response to the user-inputted pesticide combination, the system performs a relationship evaluation by combining the pesticide knowledge graph with the compatibility mapping model, obtains a compatibility relationship vector, and outputs a corresponding compatibility query report, including: The matching relationship vector is semantically decoded according to the preset structured evaluation rules to obtain the semantic encoding vector corresponding to the matching relationship vector; Based on the semantic encoding vector, a matching query text information is generated by combining it with a natural language model; The matching relationship vector is visualized, and the visualization results are combined with the matching query text information to form the matching query report.

[0068] In this embodiment, firstly, the compatibility relationship vector is semantically decoded according to preset structured evaluation rules to convert the numerical prediction vector output by the model into a semantically encoded vector with practical business meaning. The structured evaluation rules are pre-defined based on professional knowledge and practical application needs in the field of pesticide compatibility, and corresponding semantic labels or interpretations are set for the numerical range of each dimension in the compatibility relationship vector.

[0069] For example, for the chemical compatibility dimension, if the value of this dimension in the vector is in the range [0.8, 1.0], the semantic label is "fully compatible"; in the range [0.5, 0.8), the label is "basically compatible"; in the range [0.2, 0.5), the label is "minor conflict"; and in the range [0, 0.2), the label is "serious conflict". By traversing each dimension of the compatibility relationship vector and matching its value with the preset range, the semantic label corresponding to each dimension is obtained. The semantic encoding vector realizes the transformation from abstract numerical values ​​to concrete semantics, laying the foundation for the subsequent generation of natural language reports.

[0070] Secondly, based on the obtained semantic encoding vector, a natural language model is used to generate compatibility query text information. The natural language model, based on the Transformer architecture and the GPT series, takes the semantic encoding vector as input and automatically generates natural language text based on the semantic label sequence in the vector, combined with the professional expression habits and grammatical rules of pesticide compatibility. This text information describes the specific performance and evaluation results of the tested pesticide combination in various compatibility dimensions. For example, if the chemical compatibility dimension label in the semantic encoding vector is "basically compatible," the synergistic effect dimension label is "significantly synergistic," and the ecological risk dimension label is "low risk," the generated text might include: "This pesticide combination is basically compatible in chemical properties, and adverse reactions are unlikely to occur after mixing; it exhibits significant synergistic effects in terms of biological activity, effectively improving control efficacy; its potential risk to the ecological environment is assessed as low, and its use at the recommended dosage has minimal impact on non-target organisms." The generated text information not only includes evaluation conclusions for each dimension but may also supplement with corresponding explanations or suggestions as needed, such as "It is recommended to conduct small-scale compatibility testing before mixing" or "Avoid use in sensitive ecological areas."

[0071] Finally, the compatibility relationship vectors are visualized, and the visualization results are merged with the generated compatibility query text information to form a complete compatibility query report.

[0072] Specifically, visualization processing uses intuitive charts to display the numerical distribution of compatibility vectors and the comparison between various dimensions. Common visualization formats include bar charts, radar charts, heat maps, or line graphs. For example, radar charts can clearly show the scores of pesticide combinations under test in multiple dimensions such as chemical compatibility, synergistic effects, toxicological interactions, residue dissipation, and ecological impact. The value of each dimension corresponds to a vertex of the radar chart, and the area and shape of the polygon formed by connecting the lines can intuitively reflect the overall compatibility performance.

[0073] Furthermore, after visualization processing, the generated charts are integrated with the previously obtained natural language text information. Typically, the text information is located in the main body of the report, explaining the evaluation of each dimension, while the visualization charts serve as supplementary content, appended to the text or embedded in relevant paragraphs, allowing users to quickly grasp the core conclusions and data distribution. The integrated compatibility query report includes both quantitative data and intuitive chart displays, as well as textual explanations and suggestions, meeting users' needs for querying pesticide compatibility relationships.

[0074] In summary, compared to existing technologies, this application constructs a pesticide knowledge graph that integrates multi-source data, consolidating various attribute information of pesticide entities. Furthermore, by associating the multidimensional dynamic characteristics between pesticide entities with their experimentally determined compatibility relationships, a mapping model is constructed and trained, achieving accurate prediction from features to relationships and improving the intelligence, real-time performance, and reliability of pesticide compatibility queries.

[0075] In summary, the embodiments of this application have at least the following technical effects: This application provides a real-time pesticide compatibility query method based on a pesticide knowledge graph. First, pesticide entities are identified and a pesticide knowledge graph based on prior kinetic knowledge is constructed to provide data support for subsequent compatibility evaluation. Second, a compatibility mapping model is constructed and trained based on historical compatibility experimental data and the constructed pesticide knowledge graph. The original experimental data is filtered and cleaned by calculating multi-dimensional indicators such as scene similarity, region similarity, and temporal distance. A model capable of mapping the kinetic characteristics and compatibility relationships between pesticide entities is trained using machine learning methods. Finally, when responding to a user-inputted pesticide combination, the target kinetic characteristic parameters of each entity in the combination are obtained by combining the pesticide knowledge graph and input into the trained compatibility mapping model for relationship evaluation, generating a quantified compatibility vector, and then outputting a compatibility query report containing visualization results and natural language explanations. Through the above technical solutions, this application, based on the structured storage and dynamic update characteristics of knowledge graphs and the intelligent reasoning capabilities of mapping models, effectively improves the accuracy and real-time performance of pesticide compatibility queries, increases query efficiency, and thus better meets the needs of modern agricultural production for rapid and accurate acquisition of pesticide compatibility information.

[0076] Example 2, as Figure 2 As shown, based on the same inventive concept as the real-time compatibility query method based on pesticide knowledge graph provided in Embodiment 1, this application also provides a real-time compatibility query system based on pesticide knowledge graph, including: The graph recognition module 11 is used to identify pesticide entities and construct a pesticide knowledge graph based on kinetic prior knowledge. Model building module 12 is used to build and train a compatibility relationship mapping model based on historical compatibility experimental data and the pesticide knowledge graph. The report output module 13 is used to respond to the pesticide combination to be tested input by the user, combine the pesticide knowledge graph with the compatibility relationship mapping model to perform relationship evaluation, obtain the compatibility relationship vector, and output a compatibility query report accordingly.

[0077] Furthermore, in one embodiment of the application, identifying the pesticide entity includes: Identify several pesticide products in the target scenario and extract the component information of these pesticide products accordingly; The component information is summarized and several major components with pharmacological efficacy are extracted and defined as several absolute pesticide entities; By traversing multiple absolute pesticide entities, multiple adjuvant components that have physicochemical reaction characteristics with the absolute pesticide entities are identified and defined as multiple relative pesticide entities; The combined output of multiple relative pesticide entities and multiple absolute pesticide entities is the pesticide entity identification result.

[0078] Furthermore, in one embodiment of the application, a pesticide knowledge graph based on kinetic prior knowledge is constructed, including: Using the identified pesticide entity as an index, the kinetic prior knowledge related to the pesticide entity is retrieved and obtained from a predefined multi-source data source, wherein the kinetic prior knowledge includes unstructured text data and structured data; The unstructured text data and structured data are input into a pre-trained parameter parsing-normalization engine to obtain quantitative parameters of multiple pesticide entities in a preset compatibility relationship dimension. Based on the quantization parameters, a corresponding multidimensional array is generated for each pesticide entity, wherein the multidimensional array contains scalar values; Using the pesticide entity as a node and the multidimensional array as the relationship between multiple nodes, the pesticide knowledge graph is obtained by connecting multiple nodes and storing them in a structured manner.

[0079] Furthermore, in one embodiment of the application, based on historical compatibility experimental data and the pesticide knowledge graph, a compatibility relationship mapping model is constructed and trained, which includes the following steps: Obtain the original compatibility experiment dataset containing multiple original compatibility experiment items, wherein each original compatibility experiment item includes experimental crop information, experimental geographical division information, experimental time information, and experimental compatibility results; Calculate the scene similarity scalar of each original matching experiment item and the target scene in terms of crop variety, the regional similarity scalar in terms of geographical region, and the time distance scalar from the current time point; Based on preset weights, the scene similarity scalar, the region similarity scalar, and the time distance scalar are weighted and calculated to obtain the rejection probability value of each original matching experiment item. Based on the rejection probability value, the selected original matching experiment items are iteratively filtered and removed from the original matching experiment dataset; When the number of remaining original matching experiment items in the original matching experiment dataset after filtering reaches a preset threshold, the data cleaning process is terminated, and the original matching experiment dataset after filtering is output as the historical matching experiment data.

[0080] In one embodiment, the model building module 12 is specifically used for: Combining the preset compatibility relationship dimensions, the historical compatibility experimental data is analyzed to evaluate the compatibility relationship and obtain the sample compatibility relationship vector; Based on the historical compatibility experiment data, extract the sample pesticide entity set from each historical compatibility experiment item; Using the pesticide entity set as an index, the pesticide knowledge graph is traversed for entity matching, and a sample dynamic feature parameter set for each pesticide entity set is defined based on the entity matching results. By combining the sample compatibility vector with the sample dynamics feature parameter set, the compatibility mapping model is constructed and trained based on machine learning methods.

[0081] Furthermore, combining the sample compatibility vector with the sample dynamics feature parameter set, a compatibility mapping model is constructed and trained based on machine learning methods, including: Vectorize the sample dynamic feature parameter set to obtain the sample multidimensional dynamic feature vector; Establish the correspondence between the sample matching relationship vector and the sample multidimensional dynamic feature vector; Initialize the matching relationship mapping model based on machine learning, and according to the correspondence, use the multidimensional dynamic feature vector of the sample as the sample input and the matching relationship vector of the sample as the sample output to perform supervised training on the matching relationship mapping model until the preset termination constraint is met.

[0082] In one embodiment, the report output module 13 is specifically used for: Obtain the pesticide combination to be tested input by the user, and extract the corresponding pesticide entity set; Based on the pesticide knowledge graph, the target kinetic feature parameter set corresponding to the pesticide entity set to be tested is obtained by matching, wherein the target kinetic feature parameter set includes target kinetic feature parameter clusters of multiple uniquely associated pesticide entity pairs; Vectorize multiple target dynamic feature parameter clusters to obtain multiple target multidimensional dynamic feature vectors, and then concatenate the vectors. The vector concatenation result is input into the matching relationship mapping model to obtain the output matching relationship vector.

[0083] Furthermore, in response to the user-inputted pesticide combination, the relationship is evaluated by combining the pesticide knowledge graph with the compatibility mapping model to obtain a compatibility relationship vector, and a corresponding compatibility query report is output, including: The matching relationship vector is semantically decoded according to the preset structured evaluation rules to obtain the semantic encoding vector corresponding to the matching relationship vector; Based on the semantic encoding vector, a matching query text information is generated by combining it with a natural language model; The matching relationship vector is visualized, and the visualization results are combined with the matching query text information to form the matching query report.

[0084] Furthermore, the sample matching relationship vector is associated with at least one of the sample multidimensional dynamic feature vectors.

[0085] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0086] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0087] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for real-time query of pesticide compatibility relationships based on pesticide knowledge graphs, characterized in that, include: Identify pesticide entities and construct a pesticide knowledge graph based on kinetic prior knowledge; Based on historical compatibility experimental data and the pesticide knowledge graph, a compatibility relationship mapping model was constructed and trained. In response to the user-inputted pesticide combination, the system performs a relationship evaluation by combining the pesticide knowledge graph with the compatibility mapping model, obtains a compatibility relationship vector, and outputs a corresponding compatibility query report.

2. The real-time compatibility query method based on pesticide knowledge graph as described in claim 1, characterized in that, Identifying pesticide entities includes: Identify several pesticide products in the target scenario and extract the component information of these pesticide products accordingly; The component information is summarized and several major components with pharmacological efficacy are extracted and defined as several absolute pesticide entities; By traversing multiple absolute pesticide entities, multiple adjuvant components that have physicochemical reaction characteristics with the absolute pesticide entities are identified and defined as multiple relative pesticide entities; The combined output of multiple relative pesticide entities and multiple absolute pesticide entities is the pesticide entity identification result.

3. The real-time compatibility query method based on pesticide knowledge graph as described in claim 1, characterized in that, Constructing a pesticide knowledge graph based on kinetic prior knowledge, including: Using the identified pesticide entity as an index, the kinetic prior knowledge related to the pesticide entity is retrieved and obtained from a predefined multi-source data source, wherein the kinetic prior knowledge includes unstructured text data and structured data; The unstructured text data and structured data are input into a pre-trained parameter parsing-normalization engine to obtain quantitative parameters of multiple pesticide entities in a preset compatibility relationship dimension. Based on the quantization parameters, a corresponding multidimensional array is generated for each pesticide entity, wherein the multidimensional array contains scalar values; Using the pesticide entity as a node and the multidimensional array as the relationship between multiple nodes, the pesticide knowledge graph is obtained by connecting multiple nodes and storing them in a structured manner.

4. The real-time compatibility query method based on pesticide knowledge graph as described in claim 1, characterized in that, Based on historical compatibility experimental data and the aforementioned pesticide knowledge graph, a compatibility relationship mapping model was constructed and trained. Prior to this, the following steps were taken: Obtain the original compatibility experiment dataset containing multiple original compatibility experiment items, wherein each original compatibility experiment item includes experimental crop information, experimental geographical division information, experimental time information, and experimental compatibility results; Calculate the scene similarity scalar of each original matching experiment item and the target scene in terms of crop variety, the regional similarity scalar in terms of geographical region, and the time distance scalar from the current time point; Based on preset weights, the scene similarity scalar, the region similarity scalar, and the time distance scalar are weighted and calculated to obtain the rejection probability value of each original matching experiment item. Based on the rejection probability value, the selected original matching experiment items are iteratively filtered and removed from the original matching experiment dataset; When the number of remaining original matching experiment items in the original matching experiment dataset after filtering reaches a preset threshold, the data cleaning process is terminated, and the original matching experiment dataset after filtering is output as the historical matching experiment data.

5. The real-time compatibility query method based on pesticide knowledge graph as described in claim 1, characterized in that, Based on historical compatibility experimental data and the aforementioned pesticide knowledge graph, a compatibility relationship mapping model is constructed and trained, including: Combining the preset compatibility relationship dimensions, the historical compatibility experimental data is analyzed to evaluate the compatibility relationship and obtain the sample compatibility relationship vector; Based on the historical compatibility experiment data, extract the sample pesticide entity set from each historical compatibility experiment item; Using the pesticide entity set as an index, the pesticide knowledge graph is traversed for entity matching, and a sample dynamic feature parameter set for each pesticide entity set is defined based on the entity matching results. By combining the sample compatibility vector with the sample dynamics feature parameter set, the compatibility mapping model is constructed and trained based on machine learning methods.

6. The real-time compatibility query method based on pesticide knowledge graph as described in claim 5, characterized in that, Combining the sample compatibility vector with the sample dynamics feature parameter set, a compatibility mapping model is constructed and trained based on machine learning methods, including: Vectorize the sample dynamic feature parameter set to obtain the sample multidimensional dynamic feature vector; Establish the correspondence between the sample matching relationship vector and the sample multidimensional dynamic feature vector; Initialize the matching relationship mapping model based on machine learning, and according to the correspondence, use the multidimensional dynamic feature vector of the sample as the sample input and the matching relationship vector of the sample as the sample output to perform supervised training on the matching relationship mapping model until the preset termination constraint is met.

7. The real-time compatibility query method based on pesticide knowledge graph as described in claim 1, characterized in that, In response to the user-inputted pesticide combination, the system performs a relationship evaluation by combining the pesticide knowledge graph with the compatibility mapping model, obtains a compatibility relationship vector, and outputs a corresponding compatibility query report, including: Obtain the pesticide combination to be tested input by the user, and extract the corresponding pesticide entity set; Based on the pesticide knowledge graph, the target kinetic feature parameter set corresponding to the pesticide entity set to be tested is obtained by matching, wherein the target kinetic feature parameter set includes target kinetic feature parameter clusters of multiple uniquely associated pesticide entity pairs; Vectorize multiple target dynamic feature parameter clusters to obtain multiple target multidimensional dynamic feature vectors, and then concatenate the vectors. The vector concatenation result is input into the matching relationship mapping model to obtain the output matching relationship vector.

8. The real-time query method for compatibility relationships based on pesticide knowledge graphs as described in claim 1, characterized in that, In response to the user-inputted pesticide combination, the system performs a relationship evaluation by combining the pesticide knowledge graph with the compatibility mapping model, obtains a compatibility relationship vector, and outputs a corresponding compatibility query report, including: The matching relationship vector is semantically decoded according to the preset structured evaluation rules to obtain the semantic encoding vector corresponding to the matching relationship vector; Based on the semantic encoding vector, a matching query text information is generated by combining it with a natural language model; The matching relationship vector is visualized, and the visualization results are combined with the matching query text information to form the matching query report.

9. The real-time query method for compatibility relationships based on pesticide knowledge graphs as described in claim 6, characterized in that, The sample matching relationship vector is associated with at least one of the sample multidimensional dynamic feature vectors.

10. A real-time pesticide compatibility query system based on pesticide knowledge graph, characterized in that, The method for real-time querying of compatibility relationships based on pesticide knowledge graphs as described in any one of claims 1-9 includes: The knowledge graph recognition module is used to identify pesticide entities and construct a pesticide knowledge graph based on kinetic prior knowledge. The model building module is used to build and train a compatibility relationship mapping model based on historical compatibility experimental data and the pesticide knowledge graph. The report output module is used to respond to the pesticide combination to be tested input by the user, combine the pesticide knowledge graph with the compatibility relationship mapping model to evaluate the relationship, obtain the compatibility relationship vector, and output a compatibility query report accordingly.

Images