Crop target character accurate matching algorithm model

Abstract

The invention discloses a crop target character accurate matching algorithm model, and belongs to the technical field of agricultural information, and the crop target character accurate matching algorithm model comprises a data and knowledge management unit, an intelligent decision-making unit and a closed-loop optimization mechanism. Constructing a domain knowledge base by integrating genetic, environmental and agronomic data, generating genotype-environment-character combined response virtual data, and training a character prediction model by using deep learning; small sample scene parameters are optimized in combination with transfer learning, a decision report containing genetic function annotations and environment adaptation analysis is generated after multi-dimensional rule verification, and a closed-loop optimization mechanism is formed through user feedback. The character matching precision is remarkably improved, the test cost is reduced, and the method is suitable for precision agricultural breeding and planting management.

Description

Technical Field

[0001] The present invention belongs to the field of agricultural information technology, and specifically relates to an algorithm model for precise matching of target traits of crops. Background Art

[0002] In agriculture, achieving precise matching of target crop traits is crucial for increasing crop yields, improving quality, and enhancing the sustainability of agricultural production. Accurately matching crop varieties with specific growing environments and target traits can fully realize the variety's potential, reduce resource waste, and improve agricultural production efficiency. However, current methods have several shortcomings: First, data utilization is insufficient. Traditional methods often focus on a single type of data, considering only genetic information or environmental factors while ignoring the interactions between them, making it impossible to comprehensively and accurately evaluate crop trait performance. Second, they lack adaptability to complex environmental changes. Traditional methods struggle to cope with complex and changing environmental conditions such as climate change and extreme weather, making crops susceptible to environmental factors during actual cultivation, resulting in unstable yield and quality. Third, traditional variety breeding methods are long and costly. Traditional variety breeding methods require extensive field trials and time investment, making the breeding process cumbersome and costly, making it difficult to meet rapidly changing market demands. Summary of the Invention

[0003] In response to the above-mentioned pain points, the present invention provides an algorithm model for precise matching of target crop traits. By constructing a genetic-environment-trait association knowledge base and combining virtual data generation, deep learning prediction and small sample optimization, it achieves precise matching and closed-loop decision-making of target crop traits, significantly improving the efficiency of intelligent breeding.

[0004] The scheme of the present invention is as follows:

[0005] A crop target trait precise matching algorithm model, characterized by comprising the following processing units:

[0006] The data and knowledge management unit integrates genetic information, historical trait characteristics, dynamic environmental parameters, and historical agronomic data to build a domain knowledge base that includes genetic-trait causal relationships and environment-trait quantitative mapping rules:

[0007] An intelligent decision-making unit, comprising: a virtual data generation module, which generates genotype-environment-trait joint response virtual data based on the genetic-trait causal relationship and environment-trait quantitative mapping rules in the domain knowledge base;

[0008] A model training module, which uses the joint response virtual data to train a trait prediction model to learn the influence of the interaction between genotype and environment on the target trait;

[0009] The small sample fine-tuning module optimizes the parameters of the trait prediction model based on the real data of the target scenario provided by the user to generate a set of candidate varieties;

[0010] A rule verification module calls the domain knowledge base to perform genotype compliance verification and environmental adaptability verification on candidate varieties;

[0011] The decision analysis unit generates a decision report on genetic function annotation and environmental adaptability analysis for the candidate varieties that have passed the verification;

[0012] Among them, user feedback trait data drives the iterative update of the rules of the domain knowledge base and triggers the incremental learning of the model training module to form a closed-loop optimization.

[0013] Preferably, the dynamic environmental parameters integrated with the data and knowledge management unit include real-time meteorological data and soil dynamic monitoring data; the historical trait characteristics cover the yield, quality, stress resistance and other characteristics of multiple generations of crops; the historical agronomic data include planting density, fertilization schemes, irrigation strategies and other data in different years, and data cleaning, feature extraction and rule mining methods are adopted when constructing the domain knowledge base to ensure the accuracy and reliability of the genetic-trait causal relationship and the environment-trait quantitative mapping rules.

[0014] Preferably, the virtual data generation module adopts a probability distribution simulation method, combines the interaction of the genetic-trait causal relationship and the environment-trait quantitative mapping rules in the domain knowledge base, and incorporates the gene expression regulation association and dynamic environmental factors to generate virtual data reflecting the genotype-environment-trait joint response.

[0015] Preferably, the model training module uses a deep learning algorithm to train the trait prediction model, and by learning the joint response virtual data, it explores the complex influence of the interaction between genotype and environment on the target trait, thereby improving the prediction accuracy and generalization ability of the trait prediction model.

[0016] Preferably, the small sample fine-tuning module adopts a method that combines transfer learning and meta-learning to optimize the parameters of the trait prediction model based on the real data of the target scenario provided by the user; transfer learning uses the prior knowledge in the domain knowledge base to initialize the model, and meta-learning quickly adapts to the small sample data of the target scenario, thereby effectively adjusting the model parameters with a small amount of real data and generating a high-quality set of candidate varieties.

[0017] Preferably, when the rule verification module verifies the genotype compliance of candidate varieties, a multi-dimensional evaluation method is adopted, including gene dosage effect analysis and gene interaction effect analysis; when performing environmental adaptability verification, the adaptability and stability of candidate varieties under different environmental conditions are evaluated in combination with dynamic environmental parameters and specific environmental requirements of the target scenario; the rule verification module works in conjunction with the virtual data generation module, the virtual data generation module provides simulation data support for rule verification, and the result feedback of the rule verification module is used to optimize the rules for virtual data generation.

[0018] Preferably, the genetic function annotations generated by the decision analysis unit include gene function analysis, gene regulatory relationship analysis and metabolic pathway analysis to clarify the gene and molecular action principles of the target trait; environmental adaptability analysis includes suitability assessment of the target environment, environmental risk analysis and response strategy recommendations; the decision analysis unit collaborates with the model training module and the rule verification module to generate a more accurate and comprehensive decision report based on the results of model training and the feedback of rule verification.

[0019] Preferably, the user feedback trait data includes but is not limited to yield data, quality data, and stress resistance performance data after actual planting. By analyzing and mining these feedback data, the iterative update of the rules of the domain knowledge base is driven; the updated information of the domain knowledge base is transmitted to the model training module in real time, triggering its incremental learning. Incremental learning adopts online learning or small batch learning to continuously optimize the trait prediction model and form an efficient closed-loop optimization mechanism.

[0020] Compared with the prior art, the advantages of the present invention are:

[0021] (1) Data-driven precise matching: This method integrates multiple data sources, including genetic information, historical trait characteristics, dynamic environmental parameters, and historical agronomic data, to construct a comprehensive domain knowledge base. Through in-depth mining and analysis of these data, it can accurately grasp the complex interactions between crop genotypes and environmental factors, and achieve precise matching of target crop traits.

[0022] (2) Rule verification and decision support: It has a complete rule verification mechanism, including genotype compliance verification and environmental adaptability verification; through gene dosage effect analysis, gene interaction effect analysis and environmental factor fitness calculation, it can screen out the most suitable crop varieties and provide scientific decision support for agricultural production; at the same time, the decision analysis unit analyzes the screening results in detail, provides genetic function annotations, environmental fitness analysis and response strategy suggestions, to help users better understand and apply decision results;

[0023] (3) Closed-loop optimization mechanism: A closed-loop optimization mechanism of "data-model-decision-feedback" has been established, which can continuously update the knowledge base and model parameters based on the trait data fed back by users; by comparing the predicted values with the actual values, correcting the genetic-trait rules and environment-trait mapping parameters, and using the incremental learning method to optimize the model, ensuring that the model's prediction accuracy continues to improve and adapt to the ever-changing agricultural production environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of an algorithm model for precise matching of crop target traits. DETAILED DESCRIPTION

[0025] The technical solutions of the embodiments of the present invention are explained and described below, but the following embodiments are only preferred embodiments of the present invention and are not exhaustive. Based on the embodiments in the implementation manner, other embodiments obtained by those skilled in the art without creative work are all within the scope of protection of the present invention.

[0026] The proposed crop target trait precision matching algorithm model forms a complete and efficient system centered around data collection, processing, analysis, and feedback optimization, aiming to accurately match crop varieties with target traits. The entire model comprises three core components: data and knowledge management, intelligent decision-making, and decision analysis. It also incorporates a closed-loop optimization mechanism that continuously improves matching accuracy based on user feedback. The specific implementation methods will be detailed below, combining various components and modules.

[0027] Example:

[0028] 1. Data and knowledge management unit.

[0029] 1. Data integration and processing

[0030] 1-1. Data sources: Data sources cover multiple aspects, including genetic information, namely crop genotype data and gene expression profile data; historical trait characteristics, namely multi-generation field phenotypic data, including yield and quality indicators; dynamic environmental parameters, namely real-time meteorological data and soil monitoring data; and historical agronomic data, namely planting density, fertilization plan, and irrigation strategy.

[0031] 1-2. Data processing flow:

[0032] 1-2-1. Data cleaning: The Z-score method was used to process the data, remove outliers, and interpolate missing values using data from adjacent years.

[0033] 1-2-2. Feature extraction: Normalize continuous data and perform one-hot encoding on discrete data to facilitate subsequent analysis and modeling;

[0034] 1-2-3. Rule mining: Use association rule algorithms to mine the causal relationship between genetic information and traits, such as the association between certain specific genetic markers and crop stress resistance; establish mapping rules between environmental parameters and traits through regression analysis, such as the quantitative relationship between precipitation and crop yield.

[0035] Let the multi-source data set be D multi , which consists of genetic information G info , historical traits T his , dynamic environmental parameters E dyn and historical agronomic data A his Composition, namely D multi =G info ∪T his ∪E dyn ∪A his ; The cleaned data set is represented as D clean , the feature vector set obtained by feature extraction is The mined genetic-trait rule set is denoted as R G-T , the environment-trait rule set is denoted as R E-T , and then conduct systematic integration to build a structured domain knowledge base; this domain knowledge base has two core modules: genetic-trait association module: stores causal rules such as gene function annotations, gene dosage effects, and gene interaction relationships, and clarifies the molecular action path from genotype to target traits; environment-trait mapping module: stores the quantitative relationship between environmental factors and trait performance, and describes the impact of environmental changes on traits.

[0036] 2. Intelligent decision-making unit.

[0037] 1. Virtual data generation module: Based on the domain knowledge base, it uses probability distribution simulation method to generate genotype-environment-trait (GET) joint response data:

[0038] 1-1. Mathematical model: Define genotype space G = {g1, g2, …, g m} contains all possible crop genotypes, and the environmental space E={e1,e2,…,e n} contains all possible environmental combinations, the trait space T = {t1, t2, ..., t k} contains the possible values of the target trait; the joint probability distribution model is expressed as:

[0039] P(T|G,E)=f(G,E,θ)

[0040] f(G,E,θ)=μ(G,E)+σ(G,E)·∈

[0041] The conditional probability distribution P(T|G,E) represents the probability of the target trait T appearing when the crop genotype is G and the growth environment is E.

[0042] θ is a set of model rule parameters, which is extracted based on the genetic-trait causal relationship and environment-trait mapping rules in the domain knowledge base and is used to quantitatively describe what target traits different genotypes will exhibit in a specific environment;

[0043] f(G, E, θ) is a joint probability distribution model function that uses the parameter θ extracted from the domain knowledge base to calculate the most likely expression value and fluctuation range of the target trait under the combined effects of genotype G and environment E.

[0044] μ(G,E) is the expected trait value, which is the theoretical trait mean calculated based on the genetic-trait causal relationship and the environment-trait quantitative mapping in the domain knowledge base:

[0045] μ(G,E) = ∑(genetic main effect + gene interaction effect + environment main effect + G×E interaction effect)

[0046] σ(G,E) is the variance caused by environmental fluctuations, which characterizes the degree of influence of environmental uncertainty on traits and quantifies the differences in trait performance caused by environmental factor fluctuations. Assuming that the variance of environmental fluctuations is calculated based on historical data, it can be expressed in the following way: Assume that for a given genotype g i ∈G and environment e j ∈E, among N observed samples, the actual trait value of the kth sample is t ijk , the expected trait value is μ(g i ,e j ), then the environmental fluctuation variance σ(g i ,e j ) can be expressed as:

[0047]

[0048] Where i = 1, 2, ..., m, j = 1, 2, ..., n. For the entire genotype space G and the environmental space E, σ(G, E) is a matrix, each element σ(g i ,e j ) represents the variance of environmental fluctuations under a specific genotype and environment combination;

[0049] ∈ is the standard normal distribution random noise, ∈~N(0,1), simulating the random factors not covered by the knowledge base.

[0050] 1-2. Virtual data generation: randomly extract a genotype g from the genotype library i , select an environment combination e from the historical environment data j; According to the above mathematical model, the corresponding expected trait value μ(g i ,e j ) and variance σ(g i ,e j ), and then generate the virtual trait value t ij =μ(g i ,e j )+σ(g i ,e j )·∈; Repeat the above steps to generate a virtual trait dataset D containing a large number of samples virtual ={(g i ,e j ,t ij )}.

[0051] 2. Model training module: Utilizes genotype-environment-trait (GET) joint response virtual data to train trait prediction models. Through deep learning algorithms, the model explores the impact of the interaction between genotype and environment on target traits, enabling it to understand the trait expression mechanism under the combined action of the two.

[0052] 2-1. Data input and model purpose: Virtual data set D generated by virtual data generation module virtual ={(g i ,e j ,t ij )}, including genotype g i 、Environmental combination j and the corresponding virtual trait value t ij ; Build a prediction model f(g,e;θ), the goal is to make f(g,e;θ) as close to the true value t as possible ij , thereby capturing the interaction between genotype and environment;

[0053] 2-2. Deep learning implementation logic:

[0054] 2-2-1. Data preprocessing: genotype g i Perform one-hot encoding on the environmental parameter e j Normalize to form input features that can be processed by the model;

[0055] 2-2-2. Multi-layer neural network structure: Using a deep learning neural network architecture, the genotype and environmental characteristics of the pre-processed data are processed in layers:

[0056] Input layer: receives the genotype and environmental feature vectors after data preprocessing;

[0057] Hidden layer: Through multi-layer nonlinear transformation, it learns the complex influence of the interaction between genotype and environment on the target trait;

[0058] Output layer: outputs the predicted value of the target trait Used to evaluate the performance of candidate varieties in specific environments;

[0059] 2-2-3. Loss Function and Optimization: The mean square error (MSE) is used as the loss function to quantitatively measure the deviation between the model prediction value and the true value of the virtual data. The formula is:

[0060]

[0061] Among them, N is the virtual data sample size, f(g,e;θ) is the target trait value predicted by the model, and t is the true trait value; the minimum value of the loss function is solved by the gradient descent algorithm, and the gradient is The model parameters θ are iteratively updated in the direction of fastest reduction, so that the predicted value gradually approaches the true value, and finally the optimal mapping rule of the interaction between genotype and environment is learned.

[0062] 3. Small Sample Fine-tuning Module: When the target scenario has only a small amount of real data, this module uses a combination of transfer learning and meta-learning to enable the model to quickly adapt to the new scenario and generate candidate varieties that meet the requirements;

[0063] 3-1. Transfer Learning: Transfer learning uses the pre-trained model parameters θ0 output by the model training module as the initial parameters of the model in the target scenario. The pre-trained model learns the general interaction rules of genotype, environment, and trait through virtual data. After its parameters are fully transferred to the target scenario model, the risk of overfitting caused by random initialization in small sample scenarios can be avoided. This ensures that the model incorporates prior knowledge in the domain knowledge base in its initial state, providing an effective starting point for model optimization in the target scenario.

[0064] 3-2. Meta-learning:

[0065] 3-2-1. Source of small sample data: real dataset of target scenario provided by users:

[0066] D target ={(g k ,e k ,t k )}

[0067] g k : Crop genotypes in target scenarios;

[0068] e k : Real-time environmental parameters of the target scene;

[0069] t k : The actual trait observation value of the corresponding genotype in the target scenario;

[0070] 3-2-2. Calculate the target scene error:

[0071] Use the initial parameter θ0 to predict the real data and calculate the error between the predicted value and the measured value. The formula is:

[0072]

[0073] The size of this error directly reflects the accuracy of the model in the target scenario. A large error indicates that the model's current parameters have a large prediction deviation from the local data and need to be adjusted. A small error indicates that the model parameters are well adapted to the target scenario.

[0074] 3-2-3. Parameter directional adjustment: According to the error direction, adjust the parameters in the direction that reduces the error the fastest:

[0075] θ′=θ0-α·Error direction

[0076] Based on θ0, “directional fine-tuning” is performed to generate scene-specific parameters θ′;

[0077] Where α is the adjustment step size, the learning rate of the meta-learning method; through a small number of iterations, the model parameters θ′ are quickly adapted to the target scene.

[0078] 3-2-4. Generate a scenario-adaptive model: The adjusted parameter θ′ retains the general rules learned from the virtual data and incorporates the unique characteristics of the target scenario, forming a dedicated model f′(g, e; θ′) to provide accurate predictions for subsequent screening of candidate varieties.

[0079] 4. Trait rule verification module: Relying on the domain knowledge base to build a multi-dimensional evaluation system, candidate varieties are double-screened for genetic stability and environmental adaptability to ensure decision reliability.

[0080] 4-1. Verification of genotype compliance: Quantify the degree of influence of individual gene copy number on traits through gene dosage effect analysis, assign weights to different gene dosages based on historical data and calculate effect scores; evaluate the synergistic or antagonistic effects between genes through gene interaction effect analysis, and generate a comprehensive effect score based on the interaction coefficient of the gene pair. The weighted integration of the two generates a genotype compliance score. Qualified varieties are judged to be genetically stable and the trait expression is predictable.

[0081] 4-2. Environmental adaptability verification: First, use historical data to determine the weight of the impact of environmental factors such as temperature and precipitation on the target traits. For each factor, compare the actual environmental value with the optimal growth conditions of the variety. Combined with the fluctuation range of the factor, the single factor adaptability is calculated. Finally, the weighted comprehensive adaptability is obtained. The qualified varieties are judged to be able to perform stably under the target environment.

[0082] The rule verification module works in conjunction with the virtual data generation module. The virtual data generation module provides simulation data support for rule verification. The variety evaluation results output by the rule verification are fed back to the virtual data generation module to optimize its data generation rules, forming a closed-loop linkage mechanism of "data simulation-rule verification-parameter optimization".

[0083] 3. Decision analysis unit.

[0084] The decision analysis unit generates a detailed analysis report on the genetic function and environmental adaptability of the candidate varieties that have passed the verification, supporting users to understand and apply the decision results.

[0085] 1. Genetic function annotation: Analyze the biological functions of target genes using authoritative databases such as GO and KEGG to identify the metabolic pathways and molecular mechanisms involved; screen key regulatory nodes through gene expression correlation analysis to reveal the hierarchical regulatory relationships between genes; integrate metabolic pathway data to draw regulatory pathway maps related to target traits and clarify the molecular mechanism of how genotype affects traits;

[0086] 2. Environmental adaptability analysis: This step assesses the adaptability of environmental factors such as temperature and soil fertility one by one and outputs a single-factor score. This identifies high-risk factors that may lead to yield fluctuations or quality decline, and generates targeted response strategies based on agronomic knowledge.

[0087] This unit works in real time with the model training and rule verification modules to dynamically optimize the analysis content based on the model prediction results and verification feedback, ensuring the accuracy of genetic analysis and the feasibility of environmental strategies.

[0088] 4. Closed-loop optimization.

[0089] The trait data provided by users is used to drive iterative updates of the model:

[0090] 1. Knowledge base update: Compare the model's predicted values with the actual values provided by users, and based on the differences, modify the genetic-trait rules and environment-trait mapping parameters in the knowledge base to improve the accuracy of the knowledge base.

[0091] 2. Incremental learning: Using a small batch learning method, user feedback data is added to the training set to update the model parameters.

[0092] 3. Circular verification: Recalibrate the updated model to ensure that the model's prediction accuracy is continuously improved.

[0093] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A crop target trait precise matching algorithm model, characterized by: Includes the following processing units: The data and knowledge management unit integrates genetic information, historical trait characteristics, dynamic environmental parameters, and historical agronomic data to build a domain knowledge base that includes genetic-trait causal relationships and environment-trait quantitative mapping rules: An intelligent decision-making unit, comprising: a virtual data generation module, which generates genotype-environment-trait joint response virtual data based on the genetic-trait causal relationship and environment-trait quantitative mapping rules in the domain knowledge base; A model training module, which uses the joint response virtual data to train a trait prediction model to learn the influence of the interaction between genotype and environment on the target trait; The small sample fine-tuning module optimizes the parameters of the trait prediction model based on the real data of the target scenario provided by the user to generate a set of candidate varieties; A rule verification module calls the domain knowledge base to perform genotype compliance verification and environmental adaptability verification on candidate varieties; The decision analysis unit generates a decision report on genetic function annotation and environmental adaptability analysis for the candidate varieties that have passed the verification; Among them, user feedback trait data drives the iterative update of the rules of the domain knowledge base and triggers the incremental learning of the model training module to form a closed-loop optimization.

2. The crop target trait precise matching algorithm model according to claim 1, characterized in that: The dynamic environmental parameters integrated with the data and knowledge management unit include real-time meteorological data and soil dynamic monitoring data; the historical trait characteristics cover the yield, quality, stress resistance and other characteristics of multiple generations of crops; the historical agronomic data includes planting density, fertilization schemes, irrigation strategies and other data in different years, and data cleaning, feature extraction and rule mining methods are adopted when constructing the domain knowledge base to ensure the accuracy and reliability of the genetic-trait causal relationship and the environment-trait quantitative mapping rules.

3. The crop target trait precise matching algorithm model according to claim 1, characterized in that: The virtual data generation module adopts a probability distribution simulation method, combines the interaction of genetic-trait causal relationship and environment-trait quantitative mapping rules in the domain knowledge base, and incorporates gene expression regulation association and dynamic environmental factors to generate virtual data reflecting the joint response of genotype-environment-trait.

4. The crop target trait precise matching algorithm model according to claim 1, characterized in that: The model training module uses a deep learning algorithm to train the trait prediction model. By learning the joint response virtual data, it explores the complex influence of the interaction between genotype and environment on the target trait, thereby improving the prediction accuracy and generalization ability of the trait prediction model.

5. The crop target trait precise matching algorithm model according to claim 1, characterized in that: The small sample fine-tuning module uses a combination of transfer learning and meta-learning to optimize the parameters of the trait prediction model based on the real data of the target scenario provided by the user; Transfer learning uses prior knowledge in the domain knowledge base to initialize the model, while meta-learning quickly adapts to small sample data of the target scenario, thereby effectively adjusting model parameters with a small amount of real data and generating a high-quality set of candidate varieties.

6. The crop target trait precise matching algorithm model according to claim 1, characterized in that: When verifying the genotype compliance of candidate varieties, the rule verification module adopts a multi-dimensional evaluation method, including gene dosage effect analysis and gene interaction effect analysis; when verifying environmental adaptability, it combines dynamic environmental parameters and specific environmental requirements of the target scenario to evaluate the adaptability and stability of candidate varieties under different environmental conditions; the rule verification module works in conjunction with the virtual data generation module, the virtual data generation module provides simulated data support for rule verification, and the result feedback of the rule verification module is used to optimize the rules for virtual data generation.

7. The crop target trait precise matching algorithm model according to claim 1, characterized in that: The genetic function annotations generated by the decision analysis unit include gene function analysis, gene regulation relationship analysis and metabolic pathway analysis to clarify the gene and molecular action principles of the target trait; Environmental adaptability analysis includes suitability assessment of the target environment, environmental risk analysis, and response strategy recommendations; the decision analysis unit collaborates with the model training module and the rule verification module to generate more accurate and comprehensive decision reports based on the results of model training and feedback from rule verification.

8. The crop target trait precise matching algorithm model according to claim 1, characterized in that: The user feedback trait data includes but is not limited to yield data, quality data, and stress resistance performance data after actual planting. By analyzing and mining these feedback data, the rules of the domain knowledge base are iteratively updated; the updated information of the domain knowledge base is transmitted to the model training module in real time, triggering its incremental learning. Incremental learning adopts online learning or small batch learning to continuously optimize the trait prediction model and form an efficient closed-loop optimization mechanism.

Images