Rape variety environmental adaptability evaluation system based on machine learning

Abstract

The invention belongs to the technical field of agricultural information, and discloses a rape variety environmental adaptability evaluation system based on machine learning. The system is composed of a data acquisition standardization unit, a regional environment feature modeling unit, a variety phenotype and pedigree association unit, a feature screening and model training unit, a regional adaptability scoring and decision-making unit, a dynamic learning and evolution updating unit and an evaluation visualization and variety recommendation unit. By performing standardization processing on multi-source data, format and dimension differences are eliminated, and a high-quality data foundation is laid for subsequent analysis. Fusing environment and variety bilateral characteristics, and enabling the model to capture environment influence and variety heredity characteristics at the same time. Advanced methods such as ensemble learning and a time sequence neural network are adopted, and the complex relation is accurately mined. Compared with a traditional field test and the prior art, errors are greatly reduced, the adaptation degree of the variety in different areas is more accurately judged, a scientific basis is provided for planting recommendation, and the yield and quality of the oilseed rape are improved.

Description

Technical Field

[0001] The present invention belongs to the field of agricultural information technology, and specifically relates to a rapeseed variety environmental adaptability assessment system based on machine learning. Background Art

[0002] In the agricultural production system, rapeseed is an important oil crop in my country. Its planting area and yield are directly related to the security of edible oil supply. The environmental adaptability of the variety is the core factor that determines the effectiveness of rapeseed cultivation. It is specifically manifested in the variety's stress resistance, growth cycle matching and yield stability in a specific region, which has a decisive impact on regional agricultural economic benefits and resource utilization efficiency. The traditional model for assessing the environmental adaptability of rapeseed varieties is centered around field trials. By setting up multiple test plots in the target area and conducting 3-5 years of planting observations, the method records phenotypic data such as plant type development, pest and disease resistance, and yield composition, ultimately determining the variety's adaptability level based on multi-year averages. While this method can intuitively reflect the actual performance of a variety, it has significant limitations: first, the cycle is lengthy. Regional adaptability assessments of a variety often need to span multiple growth cycles, making it difficult to meet the agricultural production demand for rapid promotion of new varieties. Second, the cost is high. Multi-point trials involve multiple expenses, including land leasing, agricultural inputs, and manual observation. Third, environmental interference is significant. Natural climate fluctuations can easily distort test data, affecting the reliability of the assessment conclusions. Fourth, large-scale application is limited. Models relying on manual observations are difficult to cover the complex and diverse ecological regions across the country, and cannot provide an accurate basis for large-scale planting planning. With the development of agricultural information technology, data-driven evaluation methods have gradually emerged, but existing technologies still have many defects. There are obvious shortcomings in the comprehensiveness of data processing, the depth of feature fusion, the advancement of model construction and the dynamic nature of system updates. It is difficult to meet the high-precision, high-efficiency and high-reliability requirements of modern agriculture for the environmental adaptability assessment of rapeseed varieties. Summary of the Invention

[0003] The purpose of the present invention is to provide a rapeseed variety environmental adaptability assessment system based on machine learning to solve the problems raised in the above background technology.

[0004] To achieve the above objectives, the present invention provides the following technical solutions: a rapeseed variety environmental adaptability assessment system based on machine learning, the system comprising a data collection and standardization unit, a regional environmental feature modeling unit, a variety phenotype and pedigree association unit, a feature screening and model training unit, a regional adaptability scoring and decision-making unit, a dynamic learning and evolutionary update unit, and an assessment visualization and variety recommendation unit; Data collection and standardization unit: This unit obtains key information such as climate and soil from multiple heterogeneous data sources, processes it using the Z-Score standardization formula, completes format conversion, time series completion, and feature normalization, and forms a standard input set. Key climate information includes core factors affecting rapeseed growth, such as accumulated temperature, precipitation distribution, and sunshine duration; key soil information includes core factors such as nutrient content and texture; Regional environmental characteristics modeling unit: Receives data from the data collection and standardization unit, uses the principal component extraction model to obtain key factors, constructs vectors, and uses a clustering algorithm to divide the planting area to generate label vectors, achieving standardized expression of environmental characteristics to facilitate subsequent variety and regional matching; Variety phenotype and pedigree association unit: Receives variety phenotype and other data, generates pedigree embedding vectors through pedigree similarity quantification model, and concatenates them with environmental label vectors to construct bilateral feature sets, providing joint expression input for model training; Feature screening and model training unit: After receiving the feature sets on both sides of the connection, it uses mutual information scoring and other methods to screen key factors, inputs the integrated regression model and time series neural network, and jointly builds a prediction model. The generalization ability is improved through cross-validation and other methods. Regional adaptability scoring and decision-making unit: After receiving the prediction results from the feature screening and model training unit, it calculates the score and classifies it using the fuzzy logic membership function, outputs the recommendation level and the weight explanation of the influencing factors, and provides a basis for the decision-making of planting recommendations in specific areas; Dynamic learning and evolution update unit: Based on the regional adaptability score and the evaluation indicators of the decision unit and the newly added samples, it determines the sample distribution drift. If there is a significant change or performance degradation, it triggers incremental training to update the parameters to ensure continuous adaptation of the model. Significant changes refer to changes in the regional sample structure due to climate change, etc. The specific criteria are: the model prediction accuracy drops by more than 5% for three consecutive cycles, or the KL divergence of the sample distribution (an indicator that measures the difference between the new and old sample distributions) is greater than 0.3; Evaluation visualization and variety recommendation unit: Visualize the results and output a variety of charts to support users in obtaining targeted suggestions and references based on the target area, thus improving the practicality and usability of the system.

[0005] Preferably, the data acquisition standardization unit includes: (1) Multi-source data integration and standardization: Obtain geographic location-related data from the geographic information platform, collect crop growth information from remote sensing terminals, obtain field planting records from agricultural experimental bases, retrieve climate data from historical meteorological databases, and comprehensively collect key information such as climate parameters and soil characteristics. Use the Z-Score standardization formula to eliminate dimensional differences, convert unstructured data such as text and images into a structured format, fill in time series gaps through interpolation, complete feature normalization, and ultimately form a standard input set with a unified structure; Climate parameters include accumulated temperature, precipitation distribution, and sunshine duration; soil characteristics include nutrient content and texture; data comes from geographic location data from geographic information platforms, crop growth status information collected by remote sensing terminals, field planting records from agricultural experimental bases, and climate records from historical meteorological databases. Z-Score normalization formula: ; Where: is the original sample value; is the mean value of the variable in historical data; is the standard deviation of the variable; is the standardized sample value; Application purpose: Convert multi-source data into a unified scale with zero mean and unit variance, effectively eliminate the dimensional effects between different variables, and unify the basic format for regional modeling and model input.

[0006] Technical source: Classic statistical standardization method, meeting Articles 1 and 2 of the Appendix Requirements; (2) Data foundation to support subsequent links: The standard data processed by this unit provides high-quality and consistent data support for subsequent units. Its unified format and standardized dimensions ensure that the regional environmental characteristics modeling unit can accurately extract key factors and avoid analysis deviations due to data confusion; at the same time, it makes the feature fusion of units such as variety phenotype association more efficient, ensuring the reliability of system analysis results from the source.

[0007] Preferably, the regional environmental characteristics modeling unit includes: (1) Environmental factor extraction and regional division: After receiving standardized environmental parameter data, the principal component extraction model is used to reduce the dimensionality of massive data, screen out the core climate factors (such as accumulated temperature and precipitation distribution) and soil factors (such as nutrient content) that affect rapeseed growth, construct a multidimensional environmental variable vector, and then use the K-means clustering algorithm to divide the national planting areas according to environmental similarity to generate regional label vectors; Principal component extraction model expression: ; Where: is the nth environmental variable (e.g., annual precipitation, sunshine hours); is the nth coefficient of the principal component; is the kth principal component extracted.

[0008] Application purpose: Extract key environmental factors that describe regional heterogeneity and generate regional vector expressions to support the generation of environmental zoning labels and input into subsequent models.

[0009] Technology source: Feature compression technology in linear algebra and machine learning, consistent with formula derivation and clear path; (2) Laying the environmental foundation for variety matching: The generated regional label vector realizes the standardized and quantitative expression of the environmental characteristics of the planting area, making the originally complex regional environmental differences clearly identifiable, providing a clear basis for the subsequent matching analysis of varieties and regions, and enabling the variety phenotype and pedigree association units to accurately combine environmental information to carry out feature fusion, thereby improving the scientific nature of the matching.

[0010] Preferably, the variety phenotype and pedigree association unit includes: (1) Variety data processing and feature fusion: Receive rapeseed variety phenotypic traits (such as plant height and 1000-grain weight), stress resistance indicators (such as drought resistance and cold resistance), and genetic pedigree maps. Generate pedigree embedding vectors representing the variety's genetic characteristics through a pedigree similarity quantification model based on genetic marker similarity, and then splice them with the regional environmental label vector to construct a feature set that integrates both sides of the information. Lineage similarity quantitative model expression: ; Where: is the pedigree similarity between variety i and variety j; is the depth of the common ancestor of the two lineages (the closer the value, the smaller it is); is the maximum depth of the pedigree tree.

[0011] Application purpose: Construct a graph neural network adjacency matrix through pedigree relationships for variety embedding representation generation; so that the model input also includes genetic background information.

[0012] Technology source: Graph representation learning method, pedigree affinity measurement. Meet the requirement of clear logical chain; (2) Providing joint feature input: The bilateral feature set covers both environmental information and variety genetic information, providing comprehensive and relevant feature data for feature screening and model training units, enabling the model to capture the intrinsic connection between the environment and the variety at the same time, and more accurately explore the adaptability laws of the variety.

[0013] Preferably, the feature screening and model training unit includes: (1) Key factor screening and model construction: After receiving the bilateral feature set, the mutual information scoring function is used to measure the correlation between the features and the adaptability results. Combined with the random forest feature importance evaluation, the key factors that have a significant impact on the variety adaptability are screened out. The key factors are input into the integrated regression model such as the gradient boosting tree and the time series neural network (to capture the impact of seasonal changes). Through joint modeling, a regional adaptability prediction model is formed; Mutual information scoring function expression: ; Where: is the information gain between the input variable X and the output variable Y; is the entropy of Y; is the conditional entropy of Y under the condition of X.

[0014] Application purpose: To measure the influence of input variables (such as an environmental factor) on the prediction of variety performance, for feature selection and dimensionality reduction.

[0015] Technical source: Information theory axioms, important feature ranking indicators in machine learning; (2) Improving the generalization ability of the model: To solve the problem of model adaptation in different regions, K-fold cross-validation is used to reduce the risk of overfitting, and transfer learning is used to transfer the knowledge of the trained model to the new regional model, which greatly improves the generalization ability of the model; the optimized model provides stable and reliable output for the regional adaptability score and decision-making unit.

[0016] Preferably, the regional adaptability scoring and decision-making unit includes: (1) Adaptability scoring and grading: After receiving the prediction results output by the model training unit, the adaptability score of the variety in each region is calculated, and then the fuzzy logic membership function is used to map the score to clear grades such as strong adaptability, medium adaptability, and weak adaptability. At the same time, the recommended grade of the variety in different regions and the weighted explanation of the influence of various environmental factors (such as precipitation and soil nutrients) on adaptability are output; (2) Providing a basis for planting recommendations: The output results provide a scientific basis for planting recommendation decisions in specific areas. Decision makers can quickly judge the suitability of varieties through the recommendation level, and understand the key reasons for suitability or incompatibility by combining the weight of influencing factors, so as to select varieties in a targeted manner and effectively improve the planting success rate and yield.

[0017] Preferably, the dynamic learning and evolution updating unit includes: (1) Monitoring sample changes and model performance: Continuously track the stability evaluation indicators of the model output (such as prediction error fluctuations), analyze the distribution changes of the input samples in combination with the newly added annual sample set, and automatically trigger the model incremental training mechanism if the model prediction accuracy drops by more than 5% for three consecutive cycles, or the KL divergence of the sample distribution (an indicator that measures the difference between the new and old sample distributions) is greater than 0.3; (2) Ensure the long-term effectiveness of the model: Through incremental training, the key parameters of the model are updated in a targeted manner without the need to retrain the entire model, thereby efficiently adapting to changes brought about by regional environmental variations and the introduction of new varieties, so that the model can maintain good performance in the long term and avoid distortion of evaluation results due to environmental or variety updates.

[0018] Preferably, the evaluation visualization and variety recommendation unit includes: (1) Visualization of evaluation results: The final adaptability score, environmental factor contribution, and variety recommendation level are presented in a visual manner. The output zone heat map intuitively displays the distribution of adaptability scores in each region, the recommended variety ranking map is sorted by adaptability, and the prediction confidence interval map reflects the reliability of the results; (2) Provide targeted recommendations and references: The system supports users to input target area information and quickly generate targeted variety planting recommendations, while also providing multi-factor references (such as key influencing environmental factors), thereby greatly improving the practicality of the system and enabling agricultural practitioners to easily obtain scientific guidance, helping the rapeseed industry achieve precise planting and efficient development.

[0019] The beneficial effects of the present invention are as follows: 1. This invention standardizes multi-source data, eliminating format and dimensional differences, laying a high-quality data foundation for subsequent analysis. It then integrates both environmental and varietal characteristics, allowing the model to simultaneously capture environmental influences and varietal genetic characteristics. Advanced methods such as ensemble learning and temporal neural networks are used to accurately explore the complex relationship between the two. Compared with traditional field trials and existing technologies, this method significantly reduces errors and more accurately determines the adaptability of varieties in different regions, providing a scientific basis for planting recommendations.

[0020] 2. Through the system's dynamic learning and evolutionary update mechanism, the present invention can track changes in regional sample structure and model performance in real time. When the environment changes or new varieties are introduced, incremental training is automatically started to update parameters without the need to retrain the entire model. This allows the system to maintain good performance for a long time, overcomes the defect of existing technologies that models are difficult to adapt to changes, extends service life, reduces repeated development investment, and significantly reduces maintenance costs.

[0021] 3. The present invention converts complex evaluation results into intuitive charts such as zoning heat maps and recommended variety ranking charts through evaluation visualization and variety recommendation units, while providing targeted planting suggestions and multi-factor references, allowing users to quickly understand the evaluation results. Whether they are agricultural production decision makers or growers, they can easily apply this tool to efficiently obtain scientific guidance and promote the transformation of the rapeseed industry from traditional experience-based planting to intelligent and precise planting. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of the rapeseed variety environmental adaptability assessment system based on machine learning of the present invention. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0024] like Figure 1 As shown, an embodiment of the present invention provides a rapeseed variety environmental adaptability assessment system based on machine learning, which consists of a data collection and standardization unit, a regional environmental feature modeling unit, a variety phenotype and pedigree association unit, a feature screening and model training unit, a regional adaptability scoring and decision unit, a dynamic learning and evolutionary update unit, and an assessment visualization and variety recommendation unit; Data collection and standardization unit: This unit obtains key information such as climate and soil from multiple heterogeneous data sources, processes it using the Z-Score standardization formula, completes format conversion, time series completion, and feature normalization, and forms a standard input set. Regional environmental characteristics modeling unit: Receives data from the data collection and standardization unit, uses the principal component extraction model to obtain key factors, constructs vectors, and uses a clustering algorithm to divide the planting area to generate label vectors, achieving standardized expression of environmental characteristics to facilitate subsequent variety and regional matching; Variety phenotype and pedigree association unit: Receives variety phenotype and other data, generates pedigree embedding vectors through pedigree similarity quantification model, and concatenates them with environmental label vectors to construct bilateral feature sets, providing joint expression input for model training; Feature screening and model training unit: After receiving the feature sets on both sides of the connection, it uses mutual information scoring and other methods to screen key factors, inputs the integrated regression model and time series neural network, and jointly builds a prediction model. The generalization ability is improved through cross-validation and other methods. Regional adaptability scoring and decision-making unit: After receiving the prediction results from the feature screening and model training unit, it calculates the score and classifies it using the fuzzy logic membership function, outputs the recommendation level and the weight explanation of the influencing factors, and provides a basis for the decision-making of planting recommendations in specific areas; Dynamic Learning and Evolution Update Unit: Based on the regional adaptability score, the evaluation indicators of the decision unit, and newly added samples, it determines sample distribution drift. If the model prediction accuracy drops by more than 5% for three consecutive cycles, or the KL divergence of the sample distribution (an indicator that measures the difference between the new and old sample distributions) is greater than 0.3, it triggers incremental training to update parameters to ensure continuous model adaptation. Evaluation visualization and variety recommendation unit: Visualize the results and output a variety of charts to support users in obtaining targeted suggestions and references based on the target area, thus improving the practicality and usability of the system.

[0025] The normalized data is delivered in JSON format, including timestamp, region ID, feature name, and normalized value; The data collection standardization unit integrates various environmental and planting-related information through multiple channels, including geographic information platforms, remote sensing acquisition terminals, agricultural experimental sites, and meteorological record databases. It then uses standardization algorithms to eliminate dimensional differences in data from different sources, converting unstructured information into a unified structured format. Furthermore, through time series completion and feature normalization, a standardized input set with a unified format is finally formed. Z-Score normalization formula: ; Where: is the original sample value; is the mean value of the variable in historical data; is the standard deviation of the variable; is the standardized sample value; Standardized data processed by the data collection standardization unit provides high-quality and consistent data support for all subsequent steps in the system. Its unified format and dimensionality standards ensure the accuracy of key information extraction in units such as regional environmental characteristics modeling, reduce analytical bias caused by data confusion, and improve the efficiency of feature fusion in units such as variety phenotypic association, ensuring the reliability and stability of the entire system's analysis results from the source of the data.

[0026] The regional environmental characteristics modeling unit receives standardized environmental parameter data and uses a principal component extraction model to reduce the dimensionality of the massive data, screening out the core climate and soil factors that affect rapeseed growth, constructing a multidimensional environmental variable vector, and then using a clustering algorithm to divide the national planting areas based on environmental similarity to generate a label vector that can reflect the regional environmental characteristics. The elbow method was used to determine the optimal number of clusters for K-means clustering (usually 5-8, adjusted according to ecological differences in planting areas across the country); Principal component extraction model expression: ; Where: is the nth environmental variable (e.g., annual precipitation, sunshine hours); is the nth coefficient of the principal component; is the kth principal component extracted; The number of principal components was determined by the cumulative variance explained, and the first k principal components with cumulative variance ≥ 85% were retained (k was dynamically adjusted according to the dimension of the environmental variables); The regional label vector generated by the regional environmental characteristic modeling unit realizes the standardized and quantitative expression of the environmental characteristics of the planting area, so that the originally complex regional environmental differences can be clearly presented, providing a clear reference for the subsequent matching analysis of varieties and regions, and at the same time helping the variety phenotype and pedigree association unit to accurately integrate environmental information, effectively improving the scientificity and accuracy of the variety and region matching work.

[0027] Among them, the variety phenotype and pedigree association unit receives the phenotypic traits, stress resistance indicators and genetic pedigree map of the rapeseed variety, and generates a pedigree embedding vector representing the genetic characteristics of the variety through a pedigree similarity quantification model based on genetic marker similarity. It then splices it with the regional environmental label vector to construct a feature set that integrates the bilateral information. SNP markers are used as genetic markers, and the pedigree embedding vector is generated by graph neural network (GNN) with a dimension of 64; Lineage similarity quantitative model expression: ; Where: is the pedigree similarity between variety i and variety j; is the depth of the common ancestor of the two lineages (the closer the value, the smaller it is); is the maximum depth of the pedigree tree.

[0028] The bilateral feature set encompasses both environmental and cultivar genetic information, providing comprehensive and relevant feature data for feature screening and model training. This enables the model to capture the inherent connections between environment and cultivar, more accurately exploring cultivar adaptability patterns and laying the foundation for the effectiveness of subsequent model training.

[0029] Among them, after receiving the bilateral feature sets, the feature screening and model training unit measures the correlation between the features and the adaptability results through the mutual information scoring function. Combined with the feature importance assessment method, it screens out the key factors that significantly affect the adaptability of the variety. These key factors are input into the integrated regression model and time series neural network, and the regional adaptability prediction model is formed through joint modeling. The temporal neural network uses an LSTM structure with two hidden layers (32 neurons per layer), a ReLU activation function, and a linear activation function for the output layer. The gradient boosting tree contains 100 trees with a maximum depth of 5 and a learning rate of 0.1. The key factor screening adopts the intersection rule of 'mutual information score ≥ 0.2 and random forest feature importance ranking in the top 30%' to ensure that the screening results are both relevant and stable; The integrated regression model (gradient boosting tree) outputs the static contribution weight of environmental factors to adaptability, and the time series neural network outputs the dynamic influence coefficient within the growth cycle. The two are weighted summed (the weights are optimized by cross-validation) to form the final prediction result; Mutual information scoring function expression: ; Where: is the information gain between the input variable X and the output variable Y; is the entropy of Y; is the conditional entropy of Y under the condition of X.

[0030] The features with the top 30% mutual information scores or those with scores greater than 0.2 are retained as key factors; To address the model's adaptability across different regions, cross-validation was used to reduce the risk of overfitting, and transfer learning was used to transfer knowledge from the trained model to the new regional model, effectively improving the model's generalization capabilities. The optimized model provides stable and reliable output for the regional adaptability scoring and decision-making units, laying a solid foundation for subsequent evaluation and decision-making.

[0031] Transfer learning achieves knowledge transfer by freezing the feature extraction layer parameters of the trained model and fine-tuning only the output layer of the model for the new region. Adaptive training is then performed on a small number of labeled samples from the new region. Among them, after receiving the prediction results output by the model training unit, the regional adaptability scoring and decision-making unit calculates the adaptability score of the variety in each region, and uses the fuzzy logic membership function to map the score into clear levels such as strong adaptability, medium adaptability, and weak adaptability. At the same time, it outputs the recommended level of the variety in different regions and the weighted explanation of the impact of each environmental factor on adaptability; Strong adaptive grade membership function formula: ; Medium adaptive grade membership function formula ; Weakly adaptive hierarchical membership function formula ; Variable definition: Score the rapeseed variety's adaptability to the target region (ranging from 0 to 100, based on model predictions); Rating The degree of membership to a certain adaptation level (the value range is 0-1, the closer the value is to 1, the more it conforms to the level); The left slope of the strong adaptation interval of the trapezoidal fuzzy membership function is 0.1, the right slope is 0.1, and the left slope of the medium adaptation interval is 0.1, the right slope is 0.1, to ensure a smooth transition of the membership degree; The output provides a scientific basis for planting recommendation decisions in specific areas. Decision-makers can quickly determine the suitability of varieties based on the recommendation grade and, by combining the weights of influencing factors, identify the key reasons for suitability or incompatibility. This allows them to select targeted varieties, effectively improving planting success rates and yields, and providing practical guidance for agricultural production.

[0032] The dynamic learning and evolution update unit continuously tracks the stability evaluation indicators of the model output and analyzes the distribution changes of the input samples in combination with the newly added annual sample sets. When it identifies a significant change in the regional sample structure or a decline in model performance, it automatically triggers the model incremental training mechanism. When the model prediction accuracy drops by more than 5% for three consecutive cycles, or the KL divergence of the sample distribution is greater than 0.3, it is considered a significant change and triggers incremental training; Incremental training prioritizes updating the model output layer weights, using stochastic gradient descent optimization with a learning rate of 0.01 and 50 iterations. The convergence condition is that the validation set loss decreases ≤ 1e-3, and the feature extraction layer parameters are not updated. Through incremental training, key model parameters are updated in a targeted manner without the need to retrain the entire model, thereby efficiently adapting to changes brought about by regional environmental variations and the introduction of new varieties. This allows the model to maintain good performance over the long term, avoids distortion of evaluation results due to environmental or variety updates, and ensures continuous and reliable operation of the system.

[0033] The assessment visualization and variety recommendation unit is responsible for visualizing the final adaptability score, environmental factor contribution, and variety recommendation level. It uses a zoning heat map to intuitively display the distribution of adaptability scores in each region, uses a recommended variety ranking map to sort by adaptability, and uses a prediction confidence interval map to reflect the reliability of the results, making complex assessment information clear and easy to understand. The zoning heat map uses a blue-red gradient mapping (blue represents low scores, red represents high scores), and the score range [0,100] corresponds to the color gradient. The recommended varieties are sorted in descending order of adaptability scores, and the top 10 are displayed. The Assessment Visualization and Variety Recommendation module allows users to input target region information and quickly generate targeted recommendations for suitable varieties. These recommendations are accompanied by multi-factor references, including key environmental factors that influence variety adaptability. This feature significantly enhances the system's practicality, enabling agricultural practitioners to easily access scientific guidance and providing strong support for precision planting and efficient development of the rapeseed industry.

[0034] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0035] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A rapeseed variety environmental adaptability assessment system based on machine learning, characterized by: The system consists of a data collection and standardization unit, a regional environmental characteristics modeling unit, a variety phenotype and pedigree association unit, a feature screening and model training unit, a regional adaptability scoring and decision-making unit, a dynamic learning and evolutionary update unit, and an evaluation visualization and variety recommendation unit; Data collection and standardization unit: This unit obtains climate parameters, including accumulated temperature, precipitation distribution, sunshine duration, and soil characteristics, including nutrient content and texture, from multiple heterogeneous data sources. It then processes these data using the Z-Score standardization formula, completing format conversion, time series completion, and feature normalization to form a standard input set. Regional environmental characteristics modeling unit: Receives data from the data collection and standardization unit, uses the principal component extraction model to obtain key factors, constructs vectors, and uses a clustering algorithm to divide the planting area to generate label vectors to achieve standardized expression of environmental characteristics; Variety phenotype and pedigree association unit: receives variety phenotype data, generates pedigree embedding vectors through pedigree similarity quantification model, and concatenates them with environmental label vectors to construct bilateral feature sets; Feature screening and model training unit: After receiving the feature sets on both sides of the connection, it uses mutual information scoring to screen key factors, inputs the integrated regression model and time series neural network, and jointly builds a prediction model. The generalization ability is improved through cross-validation; Regional adaptability scoring and decision-making unit: After receiving the prediction results from the feature screening and model training unit, it calculates the score and classifies it using the fuzzy logic membership function, and outputs the recommendation level and the weight explanation of the impact factor; Dynamic learning and evolution update unit: Based on the regional adaptability score and the evaluation indicators of the decision unit and the newly added samples, it determines the sample distribution drift. If the model prediction accuracy drops by more than 5% for three consecutive cycles or the KL divergence of the sample distribution is greater than 0.3, incremental training and parameter update are triggered. Evaluation visualization and variety recommendation unit: Visualize the results and output a variety of charts to support users in obtaining targeted suggestions and references based on the target area.

2. The rapeseed variety environmental adaptability assessment system based on machine learning according to claim 1, characterized in that: The data acquisition standardization unit includes: (1) Multi-source data integration and standardization: Obtain geographic location-related data from the geographic information platform, collect crop growth information from remote sensing terminals, obtain field planting records from agricultural experimental bases, retrieve climate data from historical meteorological databases, comprehensively collect key information on climate parameters and soil characteristics, use the Z-Score standardization formula to eliminate dimensional differences, convert text and image unstructured data into structured formats, fill time series gaps through interpolation, and complete feature normalization; (2) Data foundation to support subsequent links: processed standard data with unified format and standardized dimensions ensures that the regional environmental characteristic modeling unit can accurately extract key factors, while making the feature fusion of variety phenotype association units more efficient.

3. The rapeseed variety environmental adaptability assessment system based on machine learning according to claim 1, characterized in that: The regional environmental characteristics modeling unit includes: (1) Environmental factor extraction and regional division: After receiving standardized environmental parameter data, the principal component extraction model is used to reduce the dimensionality of massive data, screen out the core climate factors and soil factors that affect rapeseed growth, construct a multidimensional environmental variable vector, and then use the K-means clustering algorithm to divide the national planting areas according to environmental similarity to generate regional label vectors; (2) Laying the environmental foundation for variety matching: The generated regional label vector standardizes and quantifies the environmental characteristics of the planting area, so that the variety phenotype and pedigree association units can be accurately combined with environmental information to carry out feature fusion.

4. The rapeseed variety environmental adaptability assessment system based on machine learning according to claim 1, characterized in that: The variety phenotype and pedigree association unit includes: (1) Variety data processing and feature fusion: After receiving the phenotypic traits, stress resistance indicators and genetic pedigree maps of rapeseed varieties, a pedigree similarity quantification model based on genetic marker similarity is used to generate a pedigree embedding vector representing the genetic characteristics of the variety. This vector is then combined with the regional environmental label vector to construct a feature set that integrates both sides of the information. (2) Providing joint feature input: The bilateral feature set covers both environmental information and variety genetic information, enabling the model to capture the intrinsic connection between the environment and the variety.

5. The rapeseed variety environmental adaptability assessment system based on machine learning according to claim 1, characterized in that: The feature screening and model training unit includes: (1) Key factor screening and model construction: After receiving the bilateral feature set, the mutual information scoring function is used to measure the correlation between the features and the adaptability results. Combined with the random forest feature importance evaluation, the key factors that have a significant impact on the variety adaptability are screened out. The key factors are input into the gradient boosting tree ensemble regression model and the time series neural network, and a regional adaptability prediction model is formed through joint modeling; (2) Improve model generalization ability: Reduce the risk of overfitting through K-fold cross-validation, and transfer the knowledge of the trained model to the new regional model through transfer learning.

6. The rapeseed variety environmental adaptability assessment system based on machine learning according to claim 1, characterized in that: The regional adaptability scoring and decision-making unit includes: (1) Adaptability scoring and grading: After receiving the prediction results output by the model training unit, the adaptability score of the variety in each region is calculated, and then the fuzzy logic membership function is used to map the score to a clear level of strong adaptability, medium adaptability, and weak adaptability. At the same time, the recommended level of the variety in different regions and the weighted explanation of the impact of each environmental factor on adaptability are output; (2) Provide a basis for planting recommendations: Decision makers can quickly judge the suitability of varieties through the recommendation level and understand the key reasons for suitability or incompatibility based on the weight of the influencing factors.

7. The rapeseed variety environmental adaptability assessment system based on machine learning according to claim 1, characterized in that: The dynamic learning and evolution updating unit includes: (1) Monitoring sample changes and model performance: Continuously track the stability evaluation indicators of the model output, analyze the distribution changes of the input samples in combination with the newly added annual sample set, and automatically trigger the model incremental training mechanism if the model prediction accuracy drops by more than 5% for three consecutive cycles or the KL divergence of the sample distribution is greater than 0.3; (2) Ensure the long-term effectiveness of the model: Through incremental training, the key parameters of the model are updated in a targeted manner without the need to retrain the entire model, so that the model can maintain good performance for a long time and avoid distortion of evaluation results due to environmental or variety updates.

8. The rapeseed variety environmental adaptability assessment system based on machine learning according to claim 1, characterized in that: The evaluation visualization and variety recommendation unit includes: (1) Visualization of evaluation results: The final adaptability score, environmental factor contribution, and variety recommendation level are presented in a visual manner. The output zone heat map intuitively displays the distribution of adaptability scores in each region, the recommended variety ranking map is sorted by adaptability, and the prediction confidence interval map reflects the reliability of the results; (2) Provide targeted recommendations and references: Support users to input target area information and quickly generate targeted variety planting recommendations, along with multi-factor references.

Images