Knowledge graph-based seedling optimization decision-making method and system

Abstract

The invention relates to the technical field of agricultural seedling optimization decision, and discloses a seedling optimization decision method and system based on a knowledge graph, and the method comprises the steps: constructing a causal graph, and recognizing the causal dependence relation between variables through a PC algorithm; carrying out anti-factual reasoning by adopting docalculus, and predicting the performance of the seedlings under the environment intervention condition; constructing a Bayesian network model, and carrying out uncertainty quantification by adopting variational reasoning; establishing a risk assessment index, and generating a seedling optimization strategy by applying a multi-objective optimization algorithm; performing comprehensive reasoning on the knowledge graph to generate a seedling optimization decision report; according to the method, by introducing the causal reasoning algorithm, the real causal relationship between the seedling characteristics and the environmental factors can be identified, and the method does not only depend on statistical correlation, so that the environmental adaptability prediction is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural seedling optimization decision-making, and more specifically, to a seedling optimization decision-making method and system based on knowledge graph. Background Art

[0002] Seedling selection is a key technical step in modern agricultural production, directly impacting crop yield, quality, and economic benefits. With the rapid development of precision agriculture and smart farming, higher requirements are being placed on the scientificity, accuracy, and reliability of seedling selection decisions.

[0003] Currently, seedling selection relies primarily on traditional statistical analysis methods and expert judgment. Existing technical solutions mainly include statistical analysis methods based on phenotypic characteristics, genetic analysis methods based on molecular markers, and field trials based on environmental adaptability. However, these existing technologies have the following technical problems: First, the existing seedling selection method is mainly based on statistical correlation analysis, which cannot accurately identify the true causal relationship between seedling characteristics and environmental factors. Traditional methods evaluate the relationship between variables by calculating correlation coefficients or regression analysis, but correlation is not equivalent to causality, which leads to insufficient accuracy in predicting the performance of seedlings under specific environmental conditions. According to statistics, the seedling selection method based on correlation analysis has a prediction error rate of more than 40% under complex environmental conditions; secondly, the existing technology lacks the scientific prediction ability of the effects of environmental intervention measures. When it is necessary to optimize seedling growth by adjusting environmental factors such as temperature, humidity, and light intensity, traditional methods cannot accurately predict the actual impact of these intervention measures on seedling performance, making it difficult to achieve coordinated seedling selection and environmental regulation. The same optimization; secondly, the uncertainty and risk in the existing seedling selection decision-making process have not been effectively quantified and managed. Decision makers cannot accurately assess the failure probability and potential losses of selecting specific seedlings, resulting in a lack of robustness in decisions and prone to major economic losses when facing environmental changes or emergencies; in addition, seedling-related phenotypic characteristic data, genotype information, environmental monitoring data and other knowledge are scattered and stored in different data sources and systems. There is a lack of a unified knowledge representation framework and reasoning mechanism, which cannot support complex multi-dimensional and multi-objective decision reasoning, limiting the comprehensiveness and accuracy of decisions; finally, existing technologies have low computational efficiency when dealing with complex decision-making scenarios involving large-scale, multi-variety, and multi-environmental factors, making it difficult to meet the needs of modern agricultural production for real-time decision support.

[0004] Therefore, there is an urgent need for a seedling selection decision-making technology that can accurately identify causal relationships, effectively quantify uncertainty, and unify knowledge representation and reasoning, so as to improve the scientificity and reliability of seedling selection, reduce agricultural production risks, and improve economic benefits. Summary of the Invention

[0005] The present invention provides a seedling optimization decision-making method and system based on knowledge graph, which solves the technical problems in related technologies that traditional methods only rely on statistical correlation, lack causal relationship identification, have insufficient environmental intervention prediction capabilities, and insufficient decision-making risk quantification.

[0006] The present invention provides a seedling optimization decision method based on a knowledge graph, comprising: A causal diagram including seedling characteristics, environmental factors and output indicators was constructed, and the PC algorithm was used to identify the causal dependencies between variables; A structural causal model was established based on the constructed causal graph, and counterfactual reasoning was performed using docalculus to predict the performance of seedlings under environmental intervention conditions. Based on the expansion of deterministic relationships in causal graphs into probability distributions, a Bayesian network model is constructed, and variational inference is used to quantify uncertainty. Based on the uncertainty quantification results, risk assessment indicators were established and a multi-objective optimization algorithm was applied to generate seedling selection strategies. Based on the results of counterfactual reasoning, uncertainty quantification results and risk assessment indicators, comprehensive reasoning is performed on the knowledge graph to generate a seedling optimization decision report.

[0007] Furthermore, the seedling characteristics include plant height, number of leaves, stem thickness and degree of root development; The environmental factors include temperature, humidity, light intensity, soil pH value and nutrient content.

[0008] Furthermore, the structural causal model represents each variable as a function of its parent node plus a noise term, and adopts a nonlinear neural network model.

[0009] Furthermore, the Bayesian network model quantifies epistemic uncertainty and random uncertainty, providing a probabilistic basis for risk assessment.

[0010] Furthermore, the risk assessment index determines a comprehensive risk value by calculating the weighted sum of the failure probability and impact degree of different decision strategies.

[0011] Furthermore, the multi-objective optimization algorithm simultaneously optimizes output maximization, cost minimization and risk minimization, and searches for the Pareto frontier through a genetic algorithm.

[0012] Furthermore, the variational inference adopts a structured variational method, which makes the variational distribution approach the true posterior distribution by iteratively optimizing the variational parameters.

[0013] Furthermore, the risk assessment considers the temporal and spatial distribution of risks and generates multiple risk scenarios through Monte Carlo simulation method.

[0014] Furthermore, the knowledge graph reasoning generates a complete decision report including selection basis, expected effects, risk assessment and confidence level, providing explainable seedling optimization recommendations.

[0015] The present invention provides a seedling optimization decision system based on a knowledge graph, which is used to execute the above-mentioned seedling optimization decision method based on a knowledge graph, including: Data collection module, used to collect seedling phenotypic characteristic data, genotype information and environmental factor data; A causal reasoning module, which builds causal graphs and performs counterfactual reasoning; Uncertainty quantification module, used to build Bayesian network models and perform variational inference; Risk assessment module, used to calculate the risk value of decision-making strategies; The decision generation module is used to perform comprehensive reasoning on the knowledge graph and generate a seedling optimization decision report.

[0016] The beneficial effects of the present invention are as follows: by introducing a causal inference algorithm, the present invention can identify the true causal relationship between seedling characteristics and environmental factors, rather than relying solely on statistical correlation, making environmental adaptability predictions more accurate, reducing the error rate in seedling selection, and providing a more scientific and reliable decision-making basis for agricultural production; The use of counterfactual reasoning models can accurately predict the effects of environmental interventions, achieving coordinated optimization of seedlings and the environment. When environmental parameters such as temperature, humidity, and light need to be adjusted, the present invention can scientifically predict the specific impact of these regulatory measures on seedling growth, providing effective environmental regulation guidance for precision agriculture. By using Bayesian networks to quantify uncertainty, the present invention can quantify the risks and uncertainties in the decision-making process, provide risk assessment information to decision makers, and make seedling selection decisions more robust and able to effectively respond to challenges brought about by environmental changes and emergencies. Based on the unified knowledge representation and reasoning algorithm of the knowledge graph, the present invention can integrate scattered seedling-related knowledge, support complex multi-dimensional decision reasoning, and provide a fully explainable decision process, thereby enhancing the credibility and acceptability of the decision. Through a multi-objective optimization algorithm, the present invention can find the optimal balance between yield maximization, cost minimization, and risk minimization, generate a Pareto-optimal seedling selection strategy, and meet the diverse needs of different users; The decision-making system constructed by the present invention can dynamically adjust decision-making strategies according to real-time environmental changes and new observation data. It has good adaptability and flexibility and is suitable for complex and changeable agricultural production environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1This is a flow chart of a seedling optimization decision-making method based on knowledge graph in the present invention; Figure 2 It is a radar chart comparing the performance of the seedling optimization method of the present invention; Figure 3 is a bar chart of the risk assessment results of different seedling varieties in the present invention; Figure 4 is a scatter plot of the relationship between the effects of environmental factors on seedling yield in the present invention; Figure 5 It is a line graph showing the trend of decision accuracy over time in the present invention; Figure 6 It is a pie chart of the distribution of seedling selection error rates in the present invention. DETAILED DESCRIPTION

[0018] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described in some examples may be combined in other examples.

[0019] At least one embodiment of the present invention discloses a seedling selection decision method based on knowledge graph, such as Figure 1 Shown, including: Step 1: construct a causal diagram including seedling characteristics, environmental factors and output indicators, and use the PC algorithm to identify the causal dependency relationship between variables; Step 1.1, forming a multi-dimensional data set; Collect seedling phenotypic characteristic data, genotype information and environmental factor data to form a multi-dimensional data set; Optionally, the phenotypic characteristic data include morphological indicators such as plant height, number of leaves, stem thickness, and root development of seedlings; the genotype information includes molecular marker data related to disease resistance and yield; and the environmental factor data include parameters such as temperature, humidity, light intensity, soil pH, and nutrient content.

[0020] Step 1.2, use the PC algorithm to analyze the conditional independence relationship between variables; Identify the causal dependencies between seedling characteristics, environmental factors, and output indicators. The PC algorithm passes the conditional independence test. To judge the variable and In a given set of conditions Is it conditionally independent under Indicates conditional independence.

[0021] in, represents the conditional mutual information function; represents the first random variable; represents the second random variable; Represents a set of condition variables; Indicates conditional symbol, i.e. "under given conditions"; Represents a value of zero. When the conditional mutual information is equal to zero, it means that the two variables are conditionally independent under given conditions.

[0022] It should be noted that the PC algorithm's specific implementation process involves first constructing a completely undirected graph, then gradually removing edges through conditional independence tests, and finally determining edge directions through V-structure identification. In the seedling selection scenario, the algorithm was able to identify a direct causal relationship between seedling height and light intensity, while eliminating the influence of soil moisture as a confounding factor.

[0023] Step 1.3: Combine domain expert knowledge to build a causal diagram: ; in, Represents a cause-effect diagram; Represents a node set, which is the seedling characteristic variable, environmental factor variable and output indicator variable; Represents a set of edges, which are the causal relationships between variables.

[0024] Step 2: Based on the constructed causal graph, a structural causal model was established and counterfactual reasoning was performed using docalculus to predict the performance of seedlings under environmental intervention conditions. Step 2.1, build a structural causal model based on the constructed causal diagram; Represent each variable as a function of its parent node plus a noise term.

[0025] In some embodiments, the structural causal model uses a nonlinear neural network model. For example, seedling yield can be expressed as: ; in, represents the seedling yield variable; represents a causal function; 、 、 represent the input variables of temperature, humidity, and variety characteristics respectively; represents the random noise term.

[0026] Step 2.2, use docalculus to perform counterfactual reasoning calculation; By formula: ; in represents the probability distribution after intervention, represents the probability function; represents the target variable; represents the conditional probability symbol; Represents a variable Perform intervention operations to make it take the value ; represents the intervention operator; represents the intervention variable; Indicates the specific value of the intervention variable; represents the summation symbol; Indicates the specific value of the confounding variable; represents a set of confounding variables.

[0027] Calculate the intervention variable The value is Time target variable The probability distribution of .

[0028] Step 2.3, using the counterfactual reasoning formula; The formula for counterfactual reasoning is: ; in, Indicates that the variable under the counterfactual condition The value of Represents the hypothetical variable The value is counterfactual conditions; Representing variables The structure function of Indicates the variable corresponding to the function; Representing variables The parent node set of Indicates the abbreviation of the parent node; Represents the set difference operator, which excludes certain elements from a set; Indicates the variables to be excluded; Representing variables The corresponding exogenous noise variable.

[0029] The specific application of the counterfactual reasoning model is as follows: when predicting the effect of adjusting the greenhouse temperature from 25°C to 28°C on the growth rate of tomato seedlings, the model can calculate the net effect of the temperature intervention on growth rate, assuming other environmental factors remain unchanged. By separating direct and indirect causal effects, the model provides accurate predictions of the intervention effect.

[0030] Step 3: Based on the expansion of the deterministic relationship in the causal graph into a probability distribution, a Bayesian network model is constructed, and variational inference is used to quantify uncertainty; Step 3.1: Expand the deterministic relationship in the causal graph into a probability distribution and construct a Bayesian network model; First, the causal graph structure constructed in step 1 is used as the skeleton of the Bayesian network. Depending on the type of variable, the system adopts different probability distributions. For discrete variables, such as seedling variety type, soil category, etc., the system constructs a complete conditional probability table; For continuous variables such as temperature, humidity, and yield, parameterized probability distribution (such as Gaussian distribution) is used to represent them.

[0031] The system uses the Bayesian estimation method to learn the parameters of these conditional probability distributions from historical data, ensuring that the model can accurately express the probabilistic dependency between variables. The constructed Bayesian network can model the uncertainty in the seedling growth process and provide a probabilistic basis for subsequent reasoning and decision-making.

[0032] Step 3.2: Use variational inference method for parameter learning and uncertainty quantification; Approximate complex posterior probability distributions by minimizing the KL divergence: ; in, represents the variational distribution, represents the variational distribution function, represents the latent variable; Represents the parameter that minimizes the objective function; represents the Kullback-Leibler divergence; The separator representing the KL divergence; represents the true posterior probability distribution, represents the probability density function, represents the observed variable.

[0033] Optionally, the variational inference algorithm employs a structured variational approach. In its implementation, the algorithm iteratively optimizes the variational parameters, gradually bringing the variational distribution closer to the true posterior distribution. For example, when estimating the uncertainty of a seedling disease resistance parameter, the algorithm outputs a probability distribution of the parameter rather than a point estimate, thereby quantifying the confidence of the prediction.

[0034] Step 3.3: Quantify epistemic uncertainty and aleatoric uncertainty in model predictions to provide a probabilistic basis for subsequent risk assessment; Two key categories of uncertainty are distinguished and quantified: Epistemic uncertainty: caused by inaccurate estimates of model parameters, missing data, or measurement errors; Stochastic uncertainty: caused by the inherent randomness of the environment and biological systems.

[0035] The uncertainty of parameter estimation is expressed using the Bayesian posterior distribution, and the epistemic uncertainty is quantified by the distribution width of the model parameters. At the same time, the random uncertainty is quantified by the degree of dispersion of the predicted distribution. The changing trend of uncertainty under different environmental conditions is further calculated, the high uncertainty areas and their sources are identified, and the potential value of new data in reducing uncertainty is evaluated, thereby providing decision makers with comprehensive and detailed uncertainty quantification information as the basis for risk assessment. Furthermore, the Bayesian network model has been specifically applied to seedling selection: it can represent the probabilistic dependencies between seedling disease resistance, environmental humidity, and disease incidence. When measurement errors exist in environmental humidity data, the model quantifies this uncertainty through probability distributions and considers its impact on the final outcome during the decision-making process. The model outputs predictions with confidence intervals, providing decision makers with risk quantification information.

[0036] Step 4: Establish risk assessment indicators based on the uncertainty quantification results and apply a multi-objective optimization algorithm to generate a seedling selection strategy; Step 4.1, establish risk assessment indicators; The formula is used to calculate the comprehensive risk value of different decision strategies: ; in, represents the comprehensive risk value; represents the summation symbol; Indicates the The probability of failure occurring, represents the failure probability function, Abbreviation for failure. Indicates the Failure scenarios; Represents the multiplication operator; Indicates the The impact of each failure scenario, The index representing the failure condition.

[0037] In some embodiments, the risk assessment algorithm further considers the temporal and spatial distribution of risk. For example, the algorithm can calculate the changing risk faced by seedlings at different growth stages, as well as the differences in risk when growing the same variety in different geographic regions. Using Monte Carlo simulation, the algorithm generates multiple possible risk scenarios, providing decision makers with a comprehensive risk analysis.

[0038] Step 4.2, apply the multi-objective optimization algorithm; Find the Pareto optimal solution between expected returns and risks and generate multiple candidate seedling optimization strategies.

[0039] It should be noted that the multi-objective optimization algorithm is specifically implemented in seedling selection as follows: it simultaneously optimizes the three objectives of yield maximization, cost minimization, and risk minimization. The algorithm uses a genetic algorithm to search the Pareto front and generate a series of non-dominated solutions. In practical application, when faced with the choice between high-yield variety A and highly resistant variety B, the algorithm can find the optimal seedling combination that strikes a balance between yield and resistance.

[0040] Step 4.3: Select the best seedling selection plan from the candidate strategies based on the user's risk preference and return requirements; First, a user risk preference model is constructed to identify the user's risk aversion level and utility function characteristics through interactive question-answering or historical decision analysis. The user's risk preference is combined with the set of candidate strategies generated in Step 4.2 to calculate the expected utility value of each strategy under the user's specific utility function. For risk-averse users, the system prioritizes seedling combinations with low risk but stable returns; for risk-neutral users, the system prioritizes strategies that maximize expected returns; and for risk-seeking users, the system may recommend new varieties with high risk and high returns. The system also considers factors such as the user's resource constraints, planting scale, and market demand to generate a personalized, actionable optimal seedling selection plan and its corresponding environmental control strategy.

[0041] Step 5: Based on the counterfactual reasoning results, uncertainty quantification results, and risk assessment indicators, comprehensive reasoning is performed on the knowledge graph to generate a seedling optimization decision report; Step 5.1: Integrate causal reasoning results, uncertainty quantification information, and risk assessment results, and perform comprehensive reasoning on the seedling knowledge graph; First, the counterfactual reasoning results from step 2 (predictions of environmental intervention effects), the uncertainty quantification information from step 3 (prediction confidence intervals), and the risk assessment results from step 4 (comprehensive risk value) are mapped onto a unified seedling knowledge graph. The system employs a weighted graph reasoning algorithm, assigning weights based on the reliability and relevance of different information sources to resolve potential information conflicts. The system also leverages the ontological relationships and semantic rules within the knowledge graph to conduct cross-domain knowledge reasoning, such as linking genotype information with environmental adaptability predictions or associating historical planting records with current decision-making scenarios. The system also considers dynamic changes throughout the seedling life cycle through temporal extensions of the knowledge graph, providing comprehensive, consistent, and temporally robust comprehensive reasoning results.

[0042] Step 5.2: Generate a complete decision report including the basis for seedling selection, expected results, risk assessment, and confidence level; Based on the results of comprehensive reasoning, a structured decision-making report is automatically generated. The report first presents the core indicators and comparative advantages of the recommended seedling plan, and details the basis for selection, including causal analysis and expert knowledge support. The report provides quantitative expected effects of environmental intervention measures, with confidence intervals and time response curves. The risk assessment section analyzes in detail the probability and impact of various risk factors, including multi-dimensional risks such as extreme weather, pest and disease outbreaks, and market fluctuations. The report also provides a multi-level assessment of decision confidence, from data quality, model reliability to prediction stability, and provides alternative plans and emergency measures for different confidence intervals. The system uses visual charts and natural language generation technology to ensure that the report content is professional, intuitive and easy to understand.

[0043] Step 5.3: Provide users with explainable seedling selection recommendations, including recommended seedling varieties, environmental control plans, and risk warnings; The decision report is converted into specific operational recommendations, including detailed parameters of recommended seedling varieties, optimal planting density and arrangement methods. The environmental control plan section provides precise control parameters in different stages, such as temperature, humidity, light and irrigation strategies at different growth stages, and gives the allowable fluctuation range of the parameters. Risk warnings include key monitoring indicators, early warning thresholds and response measures, such as early disease sign identification and prevention and control plans. The system uses "if-then" rules to present complex cause-and-effect relationships to ensure that users understand the inherent logic of decision recommendations. In addition, the system provides interactive explanation functions, allowing users to query the basis for specific decision points and explore the advantages and disadvantages of alternative plans, thereby enhancing decision transparency and user trust, and promoting the effective implementation of plans.

[0044] A seedling optimization decision system based on a knowledge graph is used to implement the above-mentioned seedling optimization decision method based on a knowledge graph, comprising: Data collection module, used to collect seedling phenotypic characteristic data, genotype information and environmental factor data; A causal reasoning module, which builds causal graphs and performs counterfactual reasoning; Uncertainty quantification module, used to build Bayesian network models and perform variational inference; Risk assessment module, used to calculate the risk value of decision-making strategies; The decision generation module is used to perform comprehensive reasoning on the knowledge graph and generate a seedling optimization decision report.

[0045] Here, the present invention provides an implementation example: A modern agricultural park needed to select from five tomato seedling varieties the one best suited to local environmental conditions for large-scale planting. The park had a comprehensive environmental monitoring system that collected real-time environmental data such as temperature, humidity, light intensity, and soil pH, as well as historical phenotypic and genotypic data for each variety.

[0046] First, the system collected phenotypic data for five tomato varieties, including indicators such as plant height, leaf number, stem diameter, and root development, as well as molecular marker information related to disease resistance and yield. It also collected environmental monitoring data from the park over the past three years, including temperature ranges, humidity fluctuations, light intensity distribution, and soil nutrient status.

[0047] Next, the PC algorithm analyzed the relationships between seedling characteristics and environmental factors, identifying a direct causal relationship between temperature and plant height, an indirect effect of humidity on final yield through its influence on leaf number, and a direct correlation between light intensity and stem diameter development. Based on these causal relationships, the system constructed a causal graph consisting of 15 nodes and 23 edges.

[0048] The counterfactual reasoning model then predicted the performance of each variety under different environmental control scenarios. When the greenhouse temperature was adjusted from the current 24 degrees Celsius to 27 degrees Celsius, the expected yield of variety A would increase by 12%, the yield of variety B would increase by 8%, and the yield of variety C would decrease by 3%.

[0049] The Bayesian network model further quantified the uncertainty of these predictions. For the yield prediction for variety A at 27 degrees Celsius, the model provided an 85% confidence interval for a 10% to 14% increase. For the yield decrease prediction for variety C, the confidence interval ranged from a 1% to a 5% decrease, with a 78% confidence level.

[0050] The risk assessment module calculated the combined risks of various selection options. Variety A had a combined risk of 0.15, primarily due to its sensitivity to high temperatures. Variety B had a risk of 0.22, primarily due to its relatively weak disease resistance. Variety C had a risk of 0.31, primarily due to its lack of yield stability.

[0051] Finally, the knowledge graph reasoning module integrates all the information and recommends selecting variety A as the main planting variety. It also suggests controlling the greenhouse temperature within the range of 26 to 27 degrees Celsius and the humidity between 65% and 70% to achieve the optimal balance between yield and risk.

[0052] like Figures 2 to 6 Shown are a radar chart comparing the performance of seedling selection methods, a bar chart of risk assessment results for different seedling varieties, a scatter plot of the impact of environmental factors on seedling yield, a line chart showing the trend of decision accuracy over time, and a pie chart showing the distribution of seedling selection error rates.

[0053] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. A seedling selection decision method based on knowledge graph, characterized in that: include: A causal diagram including seedling characteristics, environmental factors and output indicators was constructed, and the PC algorithm was used to identify the causal dependencies between variables; A structural causal model was established based on the constructed causal graph, and counterfactual reasoning was performed using docalculus to predict the performance of seedlings under environmental intervention conditions. Based on the expansion of deterministic relationships in causal graphs into probability distributions, a Bayesian network model is constructed, and variational inference is used to quantify uncertainty. Based on the uncertainty quantification results, risk assessment indicators were established and a multi-objective optimization algorithm was applied to generate seedling selection strategies. Based on the results of counterfactual reasoning, uncertainty quantification results and risk assessment indicators, comprehensive reasoning is performed on the knowledge graph to generate a seedling optimization decision report.

2. A seedling optimization decision method based on knowledge graph according to claim 1, characterized in that: The seedling characteristics include plant height, number of leaves, stem thickness and degree of root development; The environmental factors include temperature, humidity, light intensity, soil pH value and nutrient content.

3. The seedling optimization decision method based on knowledge graph according to claim 1, characterized in that: The structural causal model represents each variable as a function of its parent node plus a noise term, and adopts a nonlinear neural network model.

4. The seedling optimization decision method based on knowledge graph according to claim 1, characterized in that: The Bayesian network model quantifies epistemic uncertainty and aleatoric uncertainty, providing a probabilistic basis for risk assessment.

5. The seedling optimization decision method based on knowledge graph according to claim 1, characterized in that: The risk assessment index determines the comprehensive risk value by calculating the weighted sum of the failure probability and impact degree of different decision-making strategies.

6. The seedling optimization decision method based on knowledge graph according to claim 1, characterized in that: The multi-objective optimization algorithm simultaneously optimizes output maximization, cost minimization and risk minimization, and searches for the Pareto frontier through a genetic algorithm.

7. The seedling optimization decision method based on knowledge graph according to claim 1, characterized in that: The variational inference adopts a structured variational method and iteratively optimizes variational parameters to make the variational distribution approach the true posterior distribution.

8. The seedling optimization decision method based on knowledge graph according to claim 1, characterized in that: The risk assessment considers the temporal and spatial distribution of risks and generates multiple risk scenarios through Monte Carlo simulation method.

9. The seedling optimization decision method based on knowledge graph according to claim 1, characterized in that: The knowledge graph reasoning generates a complete decision report including selection basis, expected effects, risk assessment and confidence level, and provides explainable seedling optimization recommendations.

10. A seedling selection decision system based on knowledge graph, characterized in that: A method for implementing a seedling optimization decision-making method based on a knowledge graph according to any one of claims 1 to 9, comprising: Data collection module, used to collect seedling phenotypic characteristic data, genotype information and environmental factor data; A causal reasoning module, which builds causal graphs and performs counterfactual reasoning; Uncertainty quantification module, used to build Bayesian network models and perform variational inference; Risk assessment module, used to calculate the risk value of decision-making strategies; The decision generation module is used to perform comprehensive reasoning on the knowledge graph and generate a seedling optimization decision report.

Images