An orchard acidification soil improvement effect evaluation system based on a knowledge graph

Abstract

The application discloses a kind of orchard acidification soil improvement effect evaluation systems based on knowledge graph, it is related to agricultural science and technology field, including: data acquisition monitoring module, for collecting the multi-source data in the process of orchard acidification soil improvement, and pre-processing is formed multi-source data set;Dynamic knowledge graph construction module is based on the multi-source data after pre-processing, constructs the dynamic knowledge graph of introduction time dimension.The application can accurately capture the hysteresis effect and cumulative effect of orchard acidification soil improvement measures by constructing the dynamic knowledge graph of introduction time dimension, traditional evaluation method is mostly dependent on static cross-sectional data, it is difficult to fully reflect the dynamic change of improvement process, and the system directly shows the dynamic relationship between various elements in the soil improvement process by time series data analysis, which provides a more accurate basis for evaluating the improvement effect, and helps to adjust the improvement strategy in time, improves the improvement efficiency.

Description

Technical Field

[0001] This invention relates to the field of agricultural science and technology, specifically to a knowledge graph-based evaluation system for improving the effects of acidified soil in orchards. Background Technology

[0002] Soil acidification in orchards is a common problem, especially in areas where apple trees have been grown for a long time. Due to excessive use of chemical fertilizers, insufficient input of organic fertilizers, and natural factors, the soil pH value drops, affecting the normal growth and yield of fruit trees. Acidified soil leads to reduced soil nutrient availability and weakened microbial activity, which in turn affects the root development, nutrient absorption, and stress resistance of fruit trees, ultimately resulting in poor growth, reduced yield, and deterioration in quality. By improving acidified soil, the soil pH value can be increased, the physical structure and chemical properties of the soil can be improved, and the soil nutrient content and availability can be increased, providing a good soil environment for fruit tree growth.

[0003] In existing technologies, static cross-sectional data are often used to evaluate the effectiveness of soil improvement. However, soil improvement is a long-term dynamic process, making it difficult to capture the lag and cumulative effects of improvement measures. Therefore, the problem to be solved by this invention is how to extract the causal relationships in orchard acidified soil improvement areas, introduce a time dimension to construct a dynamic knowledge graph, and superimpose a probability model on the dynamic knowledge graph to quantify the causal strength between improvement measures and their effects in order to evaluate the effectiveness of soil improvement. To this end, a knowledge graph-based evaluation system for orchard acidified soil improvement is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a knowledge graph-based evaluation system for improving acidified soil in orchards, in order to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] A knowledge graph-based system for evaluating the effectiveness of orchard acidification soil improvement includes:

[0007] The data acquisition and monitoring module is used to collect multi-source data during the process of improving acidified soil in orchards, and to preprocess the data to form a multi-source dataset. The multi-source data is stored in a time-series database and supports timestamp marking. The multi-source data includes soil physicochemical indicators, microbial community structure, plant physiological parameters, and environmental data.

[0008] The dynamic knowledge graph construction module, based on preprocessed multi-source data, constructs a dynamic knowledge graph with a time dimension to capture the lag and cumulative effects of soil improvement measures.

[0009] The causal reasoning analysis engine is used to extract causal relationships on dynamic knowledge graphs and overlay probabilistic models to quantify the causal strength between improvement measures and their effects.

[0010] The effect evaluation module is used to comprehensively evaluate the soil improvement effect by combining the causal relationship displayed by the dynamic knowledge graph and the causal strength quantified by the probability model, and to objectively evaluate the soil improvement effect.

[0011] The optimization scheme recommendation module is used to provide targeted optimization schemes based on the effect evaluation results, combined with the actual situation of the orchard and the needs of agricultural production. These schemes include the selection of improvement measures, optimization of application amount and application time, etc., in order to improve the efficiency and effectiveness of orchard acidification soil improvement and promote the sustainable development of the orchard.

[0012] A further improvement to the technical solution of the present invention is that the data acquisition and monitoring module specifically includes:

[0013] By deploying a sensor network in the target orchard area, combined with experimental analysis, multi-source data were collected during the process of improving acidified soil in the orchard. This included soil physicochemical indicators (pH value, nutrient content, enzyme activity), microbial community structure (Alpha / Beta diversity), plant physiological parameters (photosynthetic parameters, biomass, chlorophyll content), and environmental data (temperature and humidity). All data were accompanied by timestamps to ensure temporal consistency.

[0014] The collected raw multi-source data undergoes preprocessing operations including cleaning, denoising, and standardization to eliminate outliers, unify data formats, and form a structured multi-source dataset.

[0015] The preprocessed multi-source data is stored in a time-series database according to time series, and stored hierarchically according to category, ensuring the traceability and integrity of long-term monitoring data.

[0016] A further improvement of the technical solution of the present invention is that: the dynamic knowledge graph construction module includes a graph structure definition unit and a graph construction unit;

[0017] The graph structure definition unit is used to define the entities, relationships, and time attributes of the knowledge graph, construct the structural framework of the dynamic knowledge graph, and ensure the logic and scalability of the graph.

[0018] The graph construction unit is used to construct a dynamic knowledge graph containing the time dimension based on the structure framework of the dynamic knowledge graph after centralized preprocessing of multi-source data. Through time series analysis, it captures the lag effect and cumulative effect of improvement measures and intuitively displays the dynamic relationship between various elements in the soil improvement process.

[0019] A further improvement to the technical solution of this invention lies in the following: the process of constructing the structural framework of the dynamic knowledge graph in the graph structure definition unit is as follows:

[0020] The needs for evaluating the improvement of acidified soil in orchards were analyzed, the core entities were identified, including soil, plants and amendments, and key attributes were defined for each entity, namely pH value, biomass and application time, to ensure that the entity description is complete and accurate.

[0021] Based on the actual logic of the improvement process, the relationships between entities are analyzed, the types and directions of relationships are clarified, and an entity relationship network is constructed. The types of relationships include promotion, inhibition, and influence.

[0022] By adding a time dimension to entities and relationships, determining the time dimension and its representation, and organically combining time attributes with entities and relationships, a structural framework for a dynamic knowledge graph with logicality and scalability is constructed.

[0023] A further improvement to the technical solution of this invention lies in the following: the process of capturing the lag effect and cumulative effect of the improvement measures in the map construction unit is as follows:

[0024] Based on the dynamic knowledge graph structure framework, entities, attributes and relationships are clearly defined, multi-source data are classified and organized according to the framework, time labels are assigned to each element, and a basic dataset that meets the construction requirements is formed.

[0025] Following the framework of dynamic knowledge graph structure, the prepared multi-source data is populated into the corresponding entities, attributes and relationships one by one to construct a dynamic knowledge graph, ensuring that all elements are fully presented and the time dimension runs through it;

[0026] Using time series analysis, the constructed dynamic knowledge graph is processed to capture the lag and cumulative effects of soil improvement measures. Visualization technology is then used to present the dynamic changes and interrelationships of various elements in soil improvement over time.

[0027] A further improvement of the technical solution of the present invention is that the causal reasoning analysis engine includes a causal relationship extraction unit and a probability model superposition unit;

[0028] The causal relationship extraction unit uses a causal inference algorithm based on Granger causality test to extract the causal relationship between improvement measures and soil characteristics and plant growth and development on a dynamic knowledge graph, and clarifies the specific impact path and mechanism of improvement measures on soil and plants.

[0029] The probability model overlay unit is used to overlay a Bayesian network-based probability model on the extracted causal relationship to quantify the causal strength between improvement measures and effects, and to evaluate the differences in the effects of different improvement measures.

[0030] A further improvement to the technical solution of this invention lies in the following: the process of clarifying the specific impact pathways and mechanisms of the improvement measures on soil and plants in the causal relationship extraction unit is as follows:

[0031] Extract time-series data of improvement measures and target indicators from dynamic knowledge graphs, ensure timestamp alignment, imput missing values, eliminate non-stationarity, and form a continuous sequence for causal testing;

[0032] For each pair of improvement measures and indicators, a maximum lag period is set, Granger causality under different lag orders is calculated, the P-value is determined by the F-test, and significant relationships are screened, i.e. P<0.05, to determine the causal direction and lag time.

[0033] Significant causal relationships are mapped back to the knowledge graph, the lag period and the intensity of effect are marked, the rationality of the path is verified by combining agronomic knowledge, pseudo-causality is eliminated, and finally a multi-level causal chain from improvement measures to effects is output.

[0034] A further improvement to the technical solution of the present invention is that the probability model superposition unit specifically includes:

[0035] Based on the causal relationships in the dynamic knowledge graph, a Bayesian network topology is constructed, where nodes represent improvement measures, soil indicators, and plant response variables, and edges represent verified causal directions. A score search is used, and the network structure is adjusted in combination with domain knowledge to ensure that it conforms to agronomic mechanisms.

[0036] The probability model is trained by superimposing the Bayesian network topology, the conditional probability table is calculated, the dependence strength between nodes is quantified by maximum likelihood estimation, and the uncertainty is assessed to calculate the confidence interval.

[0037] Based on the probabilistic model, probabilistic inference is performed to calculate the average causal effect of the improvement measures on the target indicators and output the quantitative evaluation results.

[0038] A further improvement to the technical solution of this invention lies in the following: the process of comprehensively evaluating the soil improvement effect in the effect evaluation module is as follows:

[0039] The causal relationships related to soil improvement are extracted from the dynamic knowledge graph, and the causal strength data between various factors quantified in the probability model are obtained. The two are then fused to form a structured evaluation dataset.

[0040] Based on the structured evaluation dataset, combined with soil improvement goals and actual needs, the sub-scores of each target indicator are calculated. The weighted scoring method is used to integrate causal relationships and causal strength to calculate the comprehensive score of soil improvement effect, thereby generating a dynamic evaluation report and displaying the evolution trend of improvement effect through a timeline chart.

[0041] A further improvement to the technical solution of this invention lies in the following: the process of providing targeted optimization solutions in the optimization solution recommendation module is as follows:

[0042] Based on the comprehensive score, key causal chain and changes in each indicator in the dynamic evaluation report, combined with the current soil acidification level, crop varieties and growth cycle of the orchard, we analyze the effectiveness and shortcomings of existing improvement measures, and clarify the specific needs of agricultural production for soil improvement in terms of yield, quality and ecology.

[0043] Based on the evaluation and analysis results, suitable improvement measures for the orchard are selected from the improvement measure library. The types, application ranges, and application time intervals of the improvement measures are determined to form a set of targeted optimization solutions for orchard technicians to refer to.

[0044] Due to the adoption of the above technical solution, the technical progress achieved by this invention compared to the prior art is as follows:

[0045] 1. This invention provides a knowledge graph-based evaluation system for improving the effects of acidified soil in orchards. By constructing a dynamic knowledge graph with a time dimension, it can accurately capture the lag and cumulative effects of soil improvement measures for acidified soil in orchards. Traditional evaluation methods often rely on static cross-sectional data, which is difficult to fully reflect the dynamic changes in the improvement process. However, this system uses time-series data analysis to intuitively display the dynamic relationships between various elements in the soil improvement process, providing a more accurate basis for evaluating the improvement effect and helping to adjust improvement strategies in a timely manner to improve improvement efficiency.

[0046] 2. This invention provides a knowledge graph-based evaluation system for improving acidified soil in orchards. Utilizing a causal inference algorithm based on Granger causality test, it extracts the causal relationships between improvement measures and soil characteristics and plant growth and development on a dynamic knowledge graph. This not only clarifies the specific impact paths and mechanisms of improvement measures on soil and plants but also quantifies the causal intensity through probabilistic model overlay, providing a scientific basis for optimizing improvement schemes. This allows orchard managers to more clearly understand the effects of improvement measures and adjust management strategies accordingly. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0048] Figure 1 This is a schematic diagram of the workflow of a knowledge graph-based orchard acidification soil improvement effect evaluation system according to the present invention.

[0049] Figure 2This is a schematic diagram of data flow in an orchard acidification soil improvement effect evaluation system based on knowledge graph, according to the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Example 1, such as Figure 1 , Figure 2 As shown, this invention provides a knowledge graph-based evaluation system for improving the effectiveness of acidified soil in orchards, comprising:

[0052] The data acquisition and monitoring module is used to collect multi-source data during the process of improving acidified soil in orchards, and to preprocess the data to form a multi-source dataset. The multi-source data is stored in a time-series database and supports timestamp marking. The multi-source data includes soil physicochemical indicators, microbial community structure, plant physiological parameters, and environmental data. Through a sensor network deployed in the target orchard area, combined with experimental analysis, multi-source data during the process of improving acidified soil in orchards is collected, including soil physicochemical indicators (pH value, nutrient content, enzyme activity), microbial community structure (Alpha / Beta diversity), plant physiological parameters (photosynthetic parameters, biomass, chlorophyll content), and environmental data (temperature and humidity). All data are timestamped to ensure time-series consistency. The collected raw multi-source data undergoes preprocessing operations including cleaning, noise reduction, and standardization to eliminate outliers, unify data formats, and form a structured multi-source dataset. The preprocessed multi-source data is stored in the time-series database according to time series, stored hierarchically by category, and the traceability and integrity of long-term monitoring data are ensured.

[0053] The specific tasks of the data acquisition and monitoring module are as follows: Within the target orchard area, a comprehensive sensor network is deployed to monitor the orchard acidification soil improvement process. Combined with laboratory analysis methods, multi-source dynamic data on the orchard acidification soil improvement process is collected in real time. Monitored indicators include soil physicochemical indicators (pH value, nutrient content, enzyme activity), microbial community structure (Alpha / Beta diversity), plant physiological parameters (photosynthetic parameters, biomass, chlorophyll content), and environmental data (temperature and humidity). All collected data are accompanied by precise timestamps to ensure temporal consistency. The system performs preprocessing operations on the large amount of noise and outliers contained in the collected raw multi-source data. Preprocessing steps include data cleaning (removing invalid data). The data undergoes various processing steps, including data normalization (e.g., data duplication), smoothing (to reduce short-term fluctuations), standardization (to unify data from different sensors), and missing value imputation (using time series interpolation or machine learning methods to fill in missing values). These steps normalize the data from different sources to make them comparable, thus forming a structured multi-source dataset. The preprocessed multi-source data is then stored in a time-series database for efficient storage and management. The time-series database employs a hierarchical storage design, storing different types of data at different levels for rapid retrieval and querying. Simultaneously, the database supports long-term monitoring of data traceability and integrity, recording the data's source, processing, and storage location information to ensure data authenticity and reliability.

[0054] The dynamic knowledge graph construction module, based on preprocessed multi-source data, constructs a dynamic knowledge graph with a time dimension to capture the lag and cumulative effects of soil improvement measures. The dynamic knowledge graph construction module includes a graph structure definition unit and a graph construction unit.

[0055] The knowledge graph structure definition unit is used to define the entities, relationships, and temporal attributes of the knowledge graph, construct the structural framework of the dynamic knowledge graph, ensure the logic and scalability of the graph, analyze the needs of orchard acidification soil improvement evaluation, identify core entities, including soil, plants, and amendments, and define key attributes for each entity, namely pH value, biomass, and application time, to ensure the completeness and accuracy of entity descriptions. Based on the actual logic of the improvement process, the associations between entities are analyzed, the relationship types and directions are clarified, and an entity relationship network is constructed. The relationship types include promotion, inhibition, and influence. A temporal dimension is added to entities and relationships, the temporal dimension and representation method are determined, and the temporal attributes are organically combined with entities and relationships to construct a structural framework for a dynamic knowledge graph with logic and scalability.

[0056] The specific work content of the diagram structure definition unit is as follows: In the evaluation of orchard acidified soil improvement, the core entities are identified, including soil, plants, and amendments. Each entity needs to define key attributes to ensure the completeness and accuracy of the description. Among these, the key attribute for soil is pH value, which directly reflects soil acidity and alkalinity and is a core indicator for measuring the degree of acidification and the effect of improvement. It has a profound impact on the availability of nutrients and microbial activity in the soil. The key attribute for plants is biomass, encompassing the total dry matter of the aboveground and underground parts, which comprehensively reflects the plant's growth status and soil fertility level. Amendments are a key factor in changing soil acidity and alkalinity; their key attribute is application time, which determines the effectiveness of the amendment in the soil. The timing of application is crucial; different application times may lead to variations in improvement effects due to factors such as soil environment and crop growth stage. Relationships between entities are defined based on the actual mechanism of soil improvement, categorized into three types: enhancement, inhibition, and influence. A "application → enhancement / inhibition" relationship exists between soil and the amendment; the relationship between plants and soil is through influence; and the relationship between the amendment and plants is indirect. All relationships must clearly define their direction and indicate their intensity (linear or non-linear). Time attributes are deeply integrated into entities and relationships. The time dimension is set according to the monitoring frequency, in days. Entity attributes must be timestamped, and relationships must record start and end times. Time representation adopts the ISO 8601 standard, supporting cross-platform compatibility. A time indexing mechanism is designed to optimize query efficiency based on time ranges, ensuring the scalability of the map in long-term monitoring.

[0057] The knowledge graph construction unit is used to centrally process preprocessed multi-source data. Following the structural framework of a dynamic knowledge graph, it constructs a dynamic knowledge graph with a time dimension. Through time series analysis, it captures the lag and cumulative effects of improvement measures, visually displaying the dynamic relationships between various elements in the soil improvement process. Based on the dynamic knowledge graph structural framework, it clarifies entities, attributes, and relationships, classifies and organizes multi-source data according to the framework, assigns time labels to each element, forming a basic dataset that meets the construction requirements. Following the dynamic knowledge graph structural framework, the prepared multi-source data is filled into the corresponding entities, attributes, and relationships one by one to construct the dynamic knowledge graph, ensuring that each element is fully presented, with the time dimension permeating it. Time series analysis methods are used to calculate the constructed dynamic knowledge graph, capturing the lag and cumulative effects of improvement measures. Visualization technology is then used to present the dynamic changes and interrelationships of various elements in soil improvement over time.

[0058] The specific tasks of the graph construction unit are as follows: Based on a predefined dynamic knowledge graph structure framework, multi-source data is classified and organized according to entities (soil, plants, amendments), attributes (pH value, biomass, application time), and relationships (enhancement, inhibition, impact). Soil entities integrate soil physicochemical and biological indicators (enzyme activity, microbial diversity); plant entities cover plant physiological parameters (photosynthetic parameters, biomass, chlorophyll content) and stress response indicators (antioxidant enzymes, malondialdehyde); amendment entities record static attributes such as type, application amount, and time. All dynamic attributes are appended with ISO8601 standard timestamps to ensure consistency in the time dimension. Relationship data needs to be labeled with the direction of action, intensity type (linear / nonlinear), and start and end times. Through structured classification and time labeling, a basic dataset supporting dynamic analysis is formed. Based on the classified and organized multi-source data, entities, attributes, and relationships are populated into a graph database (Neo4j) to construct the initial... A dynamic knowledge graph is constructed, in which entity nodes contain unique identifiers and static attributes, while dynamic attributes are stored in key-value pairs with timestamps. In addition to labeling types, relationship edges also need to embed time windows to represent lag effects. The graph construction must ensure that the time dimension is consistent throughout. After data filling, consistency verification (timestamp continuity verification) is used to ensure the integrity and logical correctness of the graph. The Granger causality test is used to analyze the temporal correlation between soil pH and soil amendment application. By examining whether historical information on amendment application has a predictive effect on the current soil pH value, a causal relationship is determined. Based on the test results, the trend of soil pH change with amendment application is analyzed, quantifying the lag period (i.e., how long after amendment application does it begin to significantly affect soil pH) or cumulative effect (i.e., the long-term effect of multiple amendment applications on soil pH). The analysis results are presented through visualization technology, dynamically displaying the evolution and interaction of each element.

[0059] The causal reasoning analysis engine is used to extract causal relationships on a dynamic knowledge graph and overlay probabilistic models to quantify the causal strength between improvement measures and their effects. The causal reasoning analysis engine includes a causal relationship extraction unit and a probabilistic model overlay unit.

[0060] The causal relationship extraction unit utilizes a causal inference algorithm based on Granger causality test to extract the causal relationship between improvement measures and soil characteristics and plant growth and development on a dynamic knowledge graph. This clarifies the specific impact paths and mechanisms of improvement measures on soil and plants. It extracts time-series data of improvement measures and target indicators from the dynamic knowledge graph, ensuring timestamp alignment, imputing missing values, eliminating non-stationarity, and forming a continuous sequence for causal testing. For each pair of improvement measures and indicators, a maximum lag period is set, Granger causality under different lag orders is calculated, and the P-value is determined through an F-test. Significant relationships (P < 0.05) are selected to determine the causal direction and lag time. Significant causal relationships are mapped back to the knowledge graph, with lag periods and intensity of action labeled. The rationality of the path is verified using agronomic knowledge, spurious causality is eliminated, and finally, a multi-level causal chain from improvement measures to effects is output.

[0061] The specific tasks of the causal relationship extraction unit are as follows: Extracting time-series data of improvement measures and target indicators from the dynamic knowledge graph, ensuring strict timestamp alignment. Specifically, this includes: filtering monitoring records of amendment application events and corresponding soil or plant indicators based on entity and relation attribute fields; unifying data with inconsistent time dimensions using linear interpolation to achieve the same frequency; imputing missing values ​​using time series interpolation; and eliminating non-stationarity through logarithmic transformation to ensure the reliability of subsequent causal tests. The processed data forms a continuous, stationary time series, stored as a structured table, with each column representing a variable and each row corresponding to an observation at a given time point. Based on the preprocessed time-series data, Granger causality tests are performed on each pair of improvement measures and indicators, setting a maximum lag period and calculating different lag orders (from 1 to the maximum lag). The F-statistic and P-value are used to test the core hypothesis that the improvement measures do not cause changes in the target index. If the P-value is less than the significance threshold (0.05), the null hypothesis is rejected, and a causal relationship is considered to exist. All significant relationships are automatically filtered, and their causal direction (measure → index or reverse), lag order, and P-value are recorded. Combinations that are not significant or whose direction does not conform to agronomic common sense are excluded. The results are output in tabular form, including the type of improver, target index, lag time, P-value, and direction of effect. Significant causal relationships are mapped back to the dynamic knowledge graph, and the relationship attributes are updated. In the original relationship edges, fields of lag time and causal strength (P-value) are added to form a causal chain with quantitative parameters. The rationality of the path is verified by combining agronomic knowledge. False causality is excluded by the domain knowledge base. Finally, a multi-level causal chain report is output to clarify the direct and indirect effects of the improvement measures.

[0062] The probabilistic model overlay unit is used to overlay a Bayesian network-based probabilistic model on the extracted causal relationships to quantify the causal strength between improvement measures and their effects, assess the differences in the effects of different improvement measures, and construct a Bayesian network topology based on causal relationships in a dynamic knowledge graph. In this topology, nodes represent improvement measures, soil indicators, and plant response variables, and edges represent verified causal directions. A score search is used, and the network structure is adjusted in conjunction with domain knowledge to ensure that it conforms to agronomic mechanisms. The probabilistic model is trained by overlaying the Bayesian network topology, and a conditional probability table is calculated. The dependence strength between nodes is quantified by maximum likelihood estimation, and uncertainty is assessed. Confidence intervals are calculated, and probabilistic inference is performed based on the probabilistic model to calculate the average causal effect of improvement measures on the target indicators and output the quantitative assessment results.

[0063] The specific tasks of the probabilistic model overlay unit are as follows: Based on the verified causal relationships in the dynamic knowledge graph, a Bayesian network topology is constructed. Nodes represent improvement measures, soil indicators, and plant response variables, covering all stages from improvement operations to soil state changes and plant growth responses. Edges represent verified causal directions, characterizing the causal connections between nodes. A scoring search method (K2 algorithm) is used to select the optimal structure from the network structure. Simultaneously, domain knowledge is incorporated to adjust the network structure, ensuring that the network structure conforms to agronomic mechanisms and avoids connections that contradict reality. This ensures that the constructed Bayesian network topology is both data-driven and aligned with professional knowledge. After constructing the Bayesian network topology, it is used as a framework for overlay training. The probabilistic model collects relevant data and calculates conditional probability tables, which describe the probability of child nodes appearing given a parent node's state. Maximum likelihood estimation is used to quantify the dependence strength between nodes. Based on observed data, the model seeks parameter values ​​that maximize the probability of data occurrences, measuring the degree of association between nodes. Simultaneously, uncertainty assessment identifies situations where the model has significant errors, ensuring the probabilistic model accurately reflects the relationships between factors. Based on the trained probabilistic model, probabilistic inference is performed to calculate the average causal effect of improvement measures on target indicators, clarifying the influence of different improvement measures on soil indicators or plant response variables. The quantitative assessment results are output, presenting the effects of improvement measures and the relative importance of each factor.

[0064] The expression for calculating the conditional probability table is as follows:

[0065]

[0066] In the formula: P(X) i |Pa(X i )) is a conditional probability table, X i For the first i Number of child node variables, Pa(X)i ) is X i The set of parent nodes, N(X) i ,Pa(X i To satisfy X i and Pa(X) i The observation frequency of the joint values, N(Pa(X) i P(X) represents the observation frequency of the parent node taking a specific value. i |Pa(X i The value of )) ranges from 0 to 1, and the closer the probability value is to 1, the stronger the dependence.

[0067] The expression for calculating the maximum likelihood estimate is as follows:

[0068]

[0069] In the formula: For maximum likelihood estimation, it means that when node X i The conditional probability value when the k-th value is taken and its parent node is taken as the j-th combination, where n is the number of observed data points, P(X i |Pa(X i ); θ) is the parameterized conditional probability, representing the value Pa(X) given the parent node. i When ) and parameter θ, node X i The probability distribution is optimized by maximizing the joint likelihood function of the observed data. The higher the likelihood value, the better the parameter θ fits the data.

[0070] The calculation expression for uncertainty assessment is as follows:

[0071]

[0072] In the formula: The conditional probability z is the maximum likelihood estimate. α / 2 The interval width reflects the accuracy of the estimate; the larger the sample size, the narrower the interval.

[0073] The formula for calculating the average causal effect is as follows:

[0074] ACE=E[Y|do(T=1)]-E[Y|do(T=0)];

[0075] Calculate the post-intervention distribution using Bayesian networks:

[0076] P(Y|do(T=1))=∑ Pa(T) P(Y|T=1,Pa(T))·P(Pa(T));

[0077] In the formula: ACE is the average causal effect, Y is the target variable, i.e., the variable whose causal effect needs to be evaluated, T is the treatment variable, do(·) is the causal intervention quantifier, which means that the treatment variable T is forcibly set to a certain value (ignoring its naturally occurring dependencies), Pa(T) is the set of parent nodes of T, i.e. the variables that directly affect the treatment variable T in the Bayesian network. When calculating the post-intervention distribution, the values ​​of the parent nodes need to be marginalized (to eliminate confounding effects); E[·] is the expected value, which represents the average value of the variable; ACE represents the expected difference between the intervention group (T=1) and the control group (T=0) on the target variable Y, reflecting the causal effect (rather than the correlation); P(Y|do(T=1)) means that by adjusting the distribution of the parent node Pa(T), the influence of confounding factors on T is eliminated, and a pure causal effect is obtained.

[0078] The effect evaluation module is used to comprehensively evaluate the soil improvement effect by combining the causal relationship displayed by the dynamic knowledge graph and the causal strength quantified by the probability model, and to objectively evaluate the soil improvement effect.

[0079] The optimization scheme recommendation module is used to provide targeted optimization schemes based on the effect evaluation results, combined with the actual situation of the orchard and the needs of agricultural production. This includes the selection of improvement measures, optimization of application amount and application time, etc., to improve the efficiency and effect of orchard acidification soil improvement and promote the sustainable development of the orchard.

[0080] Example 2, as Figure 1 , Figure 2 As shown, based on Example 1, the present invention provides a technical solution: Preferably, in the effect evaluation module, the process of comprehensively evaluating the soil improvement effect is as follows:

[0081] The causal relationships related to soil improvement are extracted from the dynamic knowledge graph, and the causal intensity data between various factors quantified in the probability model are obtained. The two are then fused to form a structured evaluation dataset. Based on the structured evaluation dataset, combined with the soil improvement goals and actual needs, the sub-scores of each target indicator are calculated. A weighted scoring method is used to integrate the causal relationships and causal intensity to calculate the comprehensive score of the soil improvement effect, and then a dynamic evaluation report is generated. The evolution trend of the improvement effect is displayed through a timeline chart.

[0082] The specific tasks of the effect evaluation module are as follows: Extracting verified causal relationships from the dynamic knowledge graph, including the association paths and temporal attributes between improvement measures, soil indicators, and plant response variables; simultaneously, obtaining quantified causal strength data from the probabilistic model; and fusing the two types of data to construct a structured evaluation dataset, ensuring that each causal chain includes: causal direction, lag time, causal strength, and timestamp; based on the structured evaluation dataset, integrating causal strength and lag period to generate sub-scores for each improvement measure on specific target indicators; then, calculating the comprehensive score of soil improvement effect through linear weighted summation using a weighted scoring method; and automatically generating a dynamic evaluation report based on the comprehensive calculation results, outputting a comprehensive score table, a list of key causal chains, and a timeline chart. The timeline chart starts from the application time of the improvement measures, overlaying the change curves of soil indicators and plant responses, and marking the significance and lag period of the causal relationship.

[0083] The formula for calculating the overall score of soil improvement effect is as follows:

[0084]

[0085] In the formula: TS is the comprehensive score of soil improvement effect, P is the total number of causal chains, and effect p This represents the target indicator node in the p-th causal chain, i.e., the plant response variable or soil target attribute ultimately affected by the improvement measures through soil indicators. It is the endpoint variable of the causal chain, used to measure the final effect of the improvement measures. p When it matches the target of the current evaluation, the value is 1; otherwise, it is 0. This filters out the causal chain that is directly related to the target indicator and avoids irrelevant chains from interfering with the total score calculation. As an indicator function, when effect p The score is 1 if the target metric is true, and 0 otherwise. Only the causal chain score related to the target metric is accumulated. `target` is the target metric to be evaluated. p The breakdown of scores for each target indicator; ACE p This represents the average causal effect of the p-th causal chain; the larger the value, the more significant the improvement effect. p 1 represents the lag time. The smaller the value, the faster the effect appears. After taking the reciprocal, a shorter lag time results in a higher score. w1 is the weighting coefficient of the average causal effect, w2 is the weighting coefficient of the lag time, and Norm(·) is the normalization function, which is used to scale each indicator to the [0,1] interval using the maximum and minimum allowable values ​​to eliminate dimensional differences and make different indicators comparable. When TS approaches 0, the improvement measures are ineffective or have a negative effect. When TS approaches 1, the improvement measures have the best effect, the average causal effect is the largest, and the lag time is the shortest.

[0086] In the optimization solution recommendation module, the process of providing targeted optimization solutions is as follows:

[0087] Based on the comprehensive score, key causal chain, and changes in various indicators in the dynamic evaluation report, combined with the current soil acidification level, crop varieties, and growth cycle of the orchard, the effectiveness and shortcomings of existing improvement measures are analyzed. The specific needs of agricultural production for soil improvement in terms of yield, quality, and ecology are clarified. Based on the evaluation and analysis results, suitable improvement measures for the orchard are selected from the improvement measure library. The types, application ranges, and application time intervals of the improvement measures are determined, forming a set of targeted optimization solutions for orchard technicians to refer to.

[0088] The specific work content of the optimization scheme recommendation module is as follows: Based on the comprehensive score of soil improvement effect, key causal chains, and changes in various indicators in the dynamic evaluation report, and combined with the actual situation of the orchard's current soil acidification level, crop varieties, and growth cycle, the module analyzes the effectiveness and shortcomings of existing improvement measures. The comprehensive score provides a direct understanding of the impact of improvement measures on the soil and crops as a whole. The key causal chains clarify the relationships and mechanisms of action among various factors, while changes in various indicators reflect the specific effects of improvement measures in different dimensions. Combined with the actual situation of the orchard, the module analyzes whether existing measures effectively improve soil acidification and whether they meet the needs of crop growth, thereby clarifying the role of agricultural production in soil improvement in yield increase, quality optimization, and ecological protection. To address specific needs in areas such as soil protection, we ensure that improvement measures effectively meet the needs of orchard development. Based on the assessment and analysis results, we select suitable improvement measures from the improvement measure database. Taking into account the causes of soil acidification, crop characteristics, and growth cycle, we preliminarily determine the types of improvement measures. At the same time, based on the degree of soil acidification and the crop's nutrient requirements, we determine the application range to avoid waste and adverse effects caused by insufficient or excessive application. Combining the crop growth cycle and local climate conditions, we plan the approximate application time range to ensure that the improvement measures play a role during the period when the crop needs them most. Through systematic analysis and scientific planning, we form a set of targeted optimization solutions to ensure that the solutions can be smoothly implemented in actual production and are available for use by orchard technicians.

[0089] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A knowledge graph-based evaluation system for improving the effect of acidified soil in orchards, characterized in that, include: The data acquisition and monitoring module is used to collect multi-source data during the process of improving acidified soil in orchards and to preprocess the data to form a multi-source dataset. The dynamic knowledge graph construction module, based on preprocessed multi-source data, constructs a dynamic knowledge graph with a time dimension to capture the lag and cumulative effects of soil improvement measures; it analyzes the needs of orchard acidified soil improvement evaluation, identifies core entities including soil, plants and amendments, and defines key attributes for each entity, namely pH value, biomass and application time. Based on the actual logic of the improvement process, the relationships between entities are analyzed, the types and directions of relationships are clarified, and an entity relationship network is constructed. The types of relationships include promotion, inhibition, and influence. Add a time dimension to entities and relationships, determine the time dimension and its representation, and organically combine time attributes with entities and relationships to construct the structural framework of a dynamic knowledge graph; Based on the dynamic knowledge graph structure framework, entities, attributes and relationships are clearly defined, multi-source data are classified and organized according to the framework, time labels are assigned to each element, and a basic dataset that meets the construction requirements is formed. Following the framework of dynamic knowledge graph structure, the prepared multi-source data is populated into the corresponding entities, attributes and relationships one by one to construct a dynamic knowledge graph; Using time series analysis, the constructed dynamic knowledge graph is processed to capture the lag and cumulative effects of soil improvement measures. Visualization technology is then used to present the dynamic changes and interrelationships of various elements in soil improvement over time. The causal reasoning analysis engine is used to extract causal relationships on a dynamic knowledge graph and overlay probabilistic models to quantify the causal strength between improvement measures and their effects. Based on the causal relationships in the dynamic knowledge graph, a Bayesian network topology is constructed, where nodes represent improvement measures, soil indicators, and plant response variables, and edges represent verified causal directions. A scoring search is used, and the network structure is adjusted in combination with domain knowledge to ensure that it conforms to agronomic mechanisms. The probability model is trained by superimposing the Bayesian network topology, the conditional probability table is calculated, the dependence strength between nodes is quantified by maximum likelihood estimation, and the uncertainty is assessed to calculate the confidence interval. Based on the probabilistic model, probabilistic inference is performed to calculate the average causal effect of the improvement measures on the target indicators and output the quantitative evaluation results. The effect evaluation module is used to comprehensively evaluate the soil improvement effect by combining the causal relationships displayed by the dynamic knowledge graph and the causal strength quantified by the probability model. The optimization solution recommendation module is used to provide targeted optimization solutions based on the effect evaluation results, combined with the actual situation of the orchard and the needs of agricultural production.

2. The knowledge graph-based evaluation system for improving acidified soil in orchards, as described in claim 1, is characterized in that: The data acquisition and monitoring module specifically includes: By deploying a sensor network in the target orchard area, and combining experimental analysis, we collected multi-source data on the process of improving acidified soil in the orchard, including soil physicochemical indicators, microbial community structure, plant physiological parameters and environmental data. All data were accompanied by timestamps. The collected raw multi-source data undergoes preprocessing operations including cleaning, denoising, and standardization to form a structured multi-source dataset. The preprocessed multi-source data is stored in a time-series database according to time series, and then stored hierarchically according to category.

3. The knowledge graph-based evaluation system for improving acidified soil in orchards, as described in claim 2, is characterized in that: The dynamic knowledge graph construction module includes a graph structure definition unit and a graph construction unit; The graph structure definition unit is used to define the entities, relationships, and time attributes of the knowledge graph, and to construct the structural framework of the dynamic knowledge graph. The graph construction unit is used to construct a dynamic knowledge graph containing the time dimension based on the multi-source data after centralized preprocessing of the multi-source data, according to the structural framework of the dynamic knowledge graph, and to capture the lag effect and cumulative effect of the improvement measures through time series analysis.

4. The knowledge graph-based evaluation system for improving acidified soil in orchards according to claim 3, characterized in that: The causal reasoning analysis engine includes a causal relationship extraction unit and a probability model overlay unit; The causal relationship extraction unit uses a causal inference algorithm based on Granger causality test to extract the causal relationship between improvement measures and soil characteristics and plant growth and development on a dynamic knowledge graph, and clarifies the specific impact path and mechanism of improvement measures on soil and plants. The probability model overlay unit is used to overlay a Bayesian network-based probability model on the extracted causal relationship to quantify the causal strength between improvement measures and effects, and to evaluate the differences in the effects of different improvement measures.

5. The knowledge graph-based evaluation system for improving acidified soil in orchards according to claim 4, characterized in that: The process of clarifying the specific impact pathways and mechanisms of the improvement measures on soil and plants in the causal relationship extraction unit is as follows: Extract time-series data of improvement measures and target indicators from dynamic knowledge graphs, align timestamps, imput missing values, and form a continuous sequence for causal testing. For each pair of improvement measures and indicators, a maximum lag period is set, Granger causality under different lag orders is calculated, the P-value is determined by the F-test, and significant relationships are screened, i.e. P<0.05, to determine the causal direction and lag time. Significant causal relationships are mapped back to the knowledge graph, the lag period and the intensity of effect are marked, the rationality of the path is verified by combining agronomic knowledge, pseudo-causality is eliminated, and finally a multi-level causal chain from improvement measures to effects is output.

6. The knowledge graph-based evaluation system for improving acidified soil in orchards according to claim 1, characterized in that: The process of comprehensively evaluating the soil improvement effect in the effect evaluation module is as follows: The causal relationships related to soil improvement are extracted from the dynamic knowledge graph, and the causal strength data between various factors quantified in the probability model are obtained. The two are then fused to form a structured evaluation dataset. Based on the structured evaluation dataset, combined with soil improvement goals and actual needs, the sub-scores of each target indicator are calculated. The weighted scoring method is used to integrate causal relationships and causal strength to calculate the comprehensive score of soil improvement effect, thereby generating a dynamic evaluation report and displaying the evolution trend of improvement effect through a timeline chart.

7. The knowledge graph-based evaluation system for improving acidified soil in orchards according to claim 6, characterized in that: The process of providing targeted optimization solutions in the optimization solution recommendation module is as follows: Based on the comprehensive score, key causal chain and changes in each indicator in the dynamic evaluation report, combined with the current soil acidification level, crop varieties and growth cycle of the orchard, the effectiveness and shortcomings of existing improvement measures are analyzed, and the specific needs of agricultural production are clarified. Based on the evaluation and analysis results, suitable improvement measures for the orchard are selected from the improvement measure library. The types, application ranges, and application time intervals of the improvement measures are determined to form a set of targeted optimization solutions for orchard technicians to refer to.

Images