A method for logical extraction in rice trait association analysis

Abstract

The present application relates to the technical field of data processing and knowledge fusion, and discloses a logical extraction method in rice trait correlation analysis, comprising: recognizing entities and predicates in multi-source heterogeneous correlation texts and constructing initial tuples, analyzing the dependency syntax of the initial tuples and stripping environmental and time sequence constraint variables, converting into discrete topological logical nodes carrying attribute interval labels and positive and negative polarity characteristics, when detecting that the attribute interval overlap degree of opposite nodes is 0, splitting out parallel feedforward branch paths in the graph database topological network, and writing the attribute interval labels into the activation control slot of the feedforward branch paths, the present application purifies the implicit conditional constraint boundary through the dependency syntax, converts data conflicts into topological parallel branches under the attribute boundary, realizes the compatible coexistence of conflict nodes, blocks semantic drift and eliminates logical deadlocks in graph merging.

Description

Technical Field

[0001] This invention relates to a logical extraction method in rice trait association analysis, belonging to the field of data processing and knowledge fusion technology. Background Technology

[0002] Currently, in relational data graph construction and data fusion, conventional path-dependent static association rule mining algorithms establish association networks between phenotypic data, gene sequences, and environmental elements. Their design relies on the globally valid monotonicity and single truth value assumption, assuming that the association polarity of subject-object pairs is globally unique. By extracting co-existing entities and static semantic features from heterogeneous texts, they automatically convert text fragments into relation edges containing positive or negative polarities, completing the construction of a global causal network. This approach, when dealing with datasets with simple boundaries and constant environments, can output a network with a clear topological hierarchy, providing a basis for retrieval and analysis. With the increasing prevalence of heterogeneous texts from cross-regional, multi-generational crop phenotypic observations... Data accumulation and the nonlinear interaction characteristics in crop development place higher demands on fusion frameworks. Crop trait expression is controlled by the synergistic perturbation of multiple factors such as external meteorological conditions, soil nutrient distribution, and planting density. Relative correlation conclusions under heterogeneous experimental conditions or different developmental stages often exhibit opposite polarities. Conventional fusion frameworks lack the means to identify the preconditions and constraints for the rules to hold. Directly and indiscriminately merging these relative causal relationships leads to logical topological overlap within the graph. This causes the system to frequently misjudge the real trait variations caused by external variables as input noise and remove them, or blindly include mutually exclusive edges, causing semantic drift and topological blockage in the causal network.

[0003] To address this conflict, an intuitive approach is to introduce a manual mapping table to supplement prior rules, or to increase the number of network parameters to accommodate local perturbations. However, static rule tables face the problem of high construction costs and limited coverage, while the large number of parameters in black-box models leads to disordered branching of causal chains. This approach, which relies on the hard stacking of external computing power, does not address the adjustment of the relational edge topology and struggles to maintain network self-consistency when facing heterogeneous information sources. Traditional methods not only have objective limitations in the static semantic alignment of underlying heterogeneous information sources, but also suffer from mechanistic deficiencies in the methods of knowledge fusion and conflict control. For example, Chinese invention patent application CN121981232A discloses a multi-source heterogeneous data fusion and knowledge graph automatic construction system, which uses frequency statistics and threshold marginalization to remove low-evidence predicates in the published text. The topological fingerprint of two-hop relationships is used to carry out entity disambiguation and conflict calibration. However, this reconstruction path based on statistical density and single truth value consensus implicitly relies on the underlying premise that conflict is noise or the global uniqueness of causal polarity. This is fundamentally mismatched with the nonlinear objective reality that crop phenotypic expression is affected by the synergistic perturbation of multidimensional external factors. Under complex rice breeding conditions, the same regulatory gene often exhibits completely opposite physiological regulatory polarities in heterogeneous experimental environments or different developmental stages. These low-frequency mutually exclusive conclusions belong to the key causal mechanism of condition specificity, rather than input noise. Forcibly implementing convergence correction or hard removal based on confidence weight in public texts will inevitably lead to the fundamental loss of the conditional divergence mechanism, and cause semantic drift and topological deadlock within the knowledge graph due to the blind pursuit of global topological self-consistency.

[0004] Therefore, the technical problem to be solved by this invention is how to automatically extract the implicit environmental and temporal constraint variables from multi-source heterogeneous text sources, complete the parallel branch reconstruction based on nonlinear divergence conditions, and establish a self-consistent topological network with compatible divergence conditions. Summary of the Invention

[0005] To address the problems in the background art, the technical solution of the present invention is as follows: A logical extraction method in rice trait association analysis, comprising: Step S1: Identify entity words and related predicates from multi-source heterogeneous associated text sequences, and construct an initial tuple containing subject, object and associated action; Step S2: Analyze the dependency syntax of the initial tuple, extract the environmental constraint variables and temporal constraint variables as topological phase boundary constraint metadata, and convert the initial tuple into a conditional logic tuple. Step S3: Based on the topological phase boundary constraint metadata, the conditional logic tuples are segmented and converted into multiple discrete topological logic nodes carrying attribute interval labels and having positive and negative polarity state quantity characteristics. Step S4: When there are multiple discrete topological logic nodes pointing to the same subject-object pair and with opposite polarities, select the opposing first discrete topological logic node and the second discrete topological logic node, and calculate the overlap degree of the attribute intervals between the first attribute interval label and the second attribute interval label of the two nodes. Step S5: If the overlap of attribute intervals is 0, retain the first discrete topological logical node and the second discrete topological logical node in place, and split two parallel feedforward branch paths in the graph database topological network. Write the first attribute interval label and the second attribute interval label into the activation control slots of the two feedforward branch paths respectively, so that the polarity mutually exclusive relational edges can coexist under different environmental attribute boundaries.

[0006] Preferably, if the overlap of attribute intervals is greater than 0 in step S5, abnormal overlap circuit breaking is performed. The abnormal overlap circuit breaking includes the following sub-steps: step S51, calculate the semantic relevance between the first discrete topology logical node and the second discrete topology logical node; step S52, if the semantic relevance reaches the preset threshold of 0.91, intercept the graph merging thread in the graph database topology network and suspend the conflicting routing branches.

[0007] Preferably, after step S52, the following sub-steps are included: Step S521, retrieving the preset engineering prior confidence table in the storage medium and comparing the source data source weights of the suspended conflicting routing branches; Step S522, determining the attenuation of the low-confidence inferior nodes using the confidence attenuation formula, which is expressed as: in, This represents the node confidence score after decay. These are the original confidence values ​​read from the engineering prior confidence table. The decay change coefficient is determined, and the value of the decay change coefficient is in the range of 0.5 to 0.9; in step S523, if the calculated node confidence value is lower than the survival threshold of 0.4, then the inferior nodes with low confidence are removed, the circuit breaker state is lifted, a globally self-consistent adaptive knowledge graph structure is constructed, and the fused self-consistent association rule instruction is output to the control terminal.

[0008] Preferably, the process of converting the initial tuple into a conditional logic tuple in step S2 includes the following sub-steps: Step S21, calling the dependency parsing operator to decouple the tree topology structure of the original sentence containing the initial tuple; Step S22, locating and extracting prepositional phrases and conditional adverbial clauses that modify related actions to define topological boundary constraint metadata; Step S23, retrieving the preset discrete ladder threshold mapping table in the storage medium, comparing the quantitative characters in the topological boundary constraint metadata, fuzzifying the continuous physical parameters into unique discrete attribute intervals, and appending and binding the discrete attribute intervals to the relational edge modifier structure of the subject-object pair.

[0009] Preferably, the process of calculating the overlap of attribute intervals in step S4 includes the following sub-steps: Step S41, obtaining the upper and lower boundaries of each unidimensional attribute interval corresponding to the first attribute interval label and the second attribute interval label in the multidimensional environmental factor space; Step S42, for each unidimensional attribute interval, comparing the lower boundary of the first attribute interval label with the upper boundary of the second attribute interval label, and comparing the upper boundary of the first attribute interval label with the lower boundary of the second attribute interval label; Step S43, if in at least one unidimensional attribute interval, the lower boundary is greater than the upper boundary, or the upper boundary is less than the lower boundary, then it is determined that the first attribute interval label and the second attribute interval label do not overlap in the multidimensional environmental factor space, and the overlap of attribute intervals is determined to be 0.

[0010] Preferably, the environmental factors in the multidimensional environmental factor space include sunshine hours, average temperature, soil moisture, diurnal temperature range, and effective accumulated temperature during the rice growth and development stage; the single-dimensional attribute interval corresponds to the numerical interval of a single environmental factor; the first attribute interval label and the second attribute interval label are generated by projecting the multidimensional environmental vector gathered by the front-end sensor onto the core physical field scale composed of temperature, humidity, and light according to the preset hedging correlation matrix.

[0011] Preferably, a time-series topology evolution dynamic compensation process is also introduced in the graph database topology network, which includes the following sub-steps after step S5: Step S53, using the historical convergence rate of the graph in the graph database topology network as the input variable, and performing weighted processing through a preset time window decay operator to calculate the topology evolution deviation for the current time period; Step S54, when the topology evolution deviation exceeds the preset safety truncation threshold, the historical constant correction rule is automatically triggered to perform step-by-step fine-tuning of the node connection weights in the graph database topology network so that the subsequent reconstructed parallel branches adaptively fit the current resting state.

[0012] Preferably, the method for determining the activation control slot threshold of the two feedforward branch paths in step S5 includes retrieving the initial environmental attributes of historically existing topological nodes in the application scenario data source from the graph database topological network, reading the historical environmental factor range corresponding to the initial environmental attributes, and using the historical environmental factor range as the reference benchmark for activation control slots, and establishing a bidirectional dependency association constraint relationship in situ in the network topology.

[0013] Preferably, the causal reasoning constraint determination after the reconstruction of the feedforward branch path in step S5 includes the following sub-steps: Step S55, when the graph database topology network receives subsequent new input data, extract the real-time environment attribute parameters carried in the new input data; Step S56, perform lossless matching between the real-time environment attribute parameters and the first attribute interval label and the second attribute interval label in the activation control slot of the two feedforward branch paths; Step S57, if the real-time environment attribute parameters fall within the attribute interval of one of the feedforward branch paths, activate the corresponding feedforward branch path to perform causal reasoning, and lock the reasoning state of the other feedforward branch path to maintain the logical self-consistency of the global knowledge graph.

[0014] Preferably, after step S523, which removes the inferior nodes with low confidence and retains the superior branches, the following steps are included: Step S524, based on the topology feedback results of the successful merging, the confidence scores of each association rule in the engineering prior confidence table are adjusted by gain or updated by attenuation, so that the engineering prior confidence table adaptively tracks the noise distribution changes of multi-source heterogeneous data sources.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. In the logical extraction of rice trait association analysis, the sentence containing the initial tuple is analyzed by logical transformation unit. Dependency syntax is used to purify the prepositional phrases and conditional adverbs that modify the action predicate. Environmental constraint variables and temporal constraint variables are stripped as topological phase boundary constraint metadata. The continuous values ​​of physical parameters are converted into discrete attribute interval labels and directly appended to the relation edge structure of the subject-object pair, thereby converting the initial tuple into a conditional logical fragment tuple. This processing path reverses the situation in conventional association network construction where heterogeneous conclusions are attributed to absolute truth due to the lack of premise constraints. It makes the condition-specific boundaries implicit in the text explicit and adaptively blocks logical conflicts caused by temporal or environmental interference in the early stage of graph merging.

[0016] 2. The topology reorganization unit receives conditional logic fragment tuples and, based on the discrete attribute interval boundaries defined by the topological phase boundary constraint metadata, segments the conditional logic fragment tuples within the virtual topology space, mapping them to discrete logic topology nodes with physically meaningful state labels, and configuring state variable features with exclusive positive and negative polarities. This path cuts off the channel that directly translates features into continuous mapping functions, and completely transforms environmental variables into independent state nodes through conditional control flow. This not only establishes a causal chain between variables and relational topology, but also provides discrete control entities for subsequent branch determination, avoiding the problem of directly erasing the real correlation rules as noise when the polarity is reversed due to environmental heterogeneity in conventional graphs.

[0017] 3. When the conflict arbitration unit detects that discrete logical topology nodes pointing to the same subject-object pair have state variables with opposite polarities, it intercepts the conventional hard removal action based on confidence and triggers the dynamic disturbance phase boundary feedforward split arbitration mechanism. By calculating the overlap between the discrete attribute interval labels of the two nodes, when the overlap is equal to 0, the two logical nodes with opposite polarities are retained in place. The scheduling unit reconstructs parallel branches in the topology network and writes the non-overlapping discrete attribute interval labels into the activation gate slots of the two parallel routing branches respectively. This discrete event-driven branch splitting enables mutually exclusive polarity relation edges to coexist under different phase boundary boundaries, thereby eliminating the risk of logical deadlock that occurs frequently during graph merging based on the system condition divergence compatibility capability. Attached Figure Description

[0018] Figure 1 This is a flowchart of the conditional logic tuple conversion and parallel branch splitting process of the present invention; Figure 2 This is a structural diagram of the conditional logic tuple conversion and conflict branch processing of the present invention.

[0019] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0021] A logical extraction method in rice trait association analysis includes: Step S1: Identify entity words and related predicates from multi-source heterogeneous associated text sequences, and construct an initial tuple containing subject, object and associated action; Step S2: Analyze the dependency syntax of the initial tuple, extract the environmental constraint variables and temporal constraint variables as topological phase boundary constraint metadata, and convert the initial tuple into a conditional logic tuple. Step S3: Based on the topological phase boundary constraint metadata, the conditional logic tuples are segmented and converted into multiple discrete topological logic nodes carrying attribute interval labels and having positive and negative polarity state quantity characteristics. Step S4: When there are multiple discrete topological logic nodes pointing to the same subject-object pair and with opposite polarities, select the opposing first discrete topological logic node and the second discrete topological logic node, and calculate the overlap degree of the attribute intervals between the first attribute interval label and the second attribute interval label of the two nodes. Step S5: If the overlap of attribute intervals is 0, retain the first discrete topological logical node and the second discrete topological logical node in place, and split two parallel feedforward branch paths in the graph database topological network. Write the first attribute interval label and the second attribute interval label into the activation control slots of the two feedforward branch paths respectively, so that the polarity mutually exclusive relational edges can coexist under different environmental attribute boundaries.

[0022] Preferably, if the overlap of attribute intervals is greater than 0 in step S5, abnormal overlap circuit breaking is performed. The abnormal overlap circuit breaking includes the following sub-steps: step S51, calculate the semantic relevance between the first discrete topology logical node and the second discrete topology logical node; step S52, if the semantic relevance reaches the preset threshold of 0.91, intercept the graph merging thread in the graph database topology network and suspend the conflicting routing branches.

[0023] Preferably, after step S52, the following sub-steps are included: Step S521, retrieving the preset engineering prior confidence table in the storage medium and comparing the source data source weights of the suspended conflicting routing branches; Step S522, determining the attenuation of the low-confidence inferior nodes using the confidence attenuation formula, which is expressed as: in, This represents the node confidence score after decay. These are the original confidence values ​​read from the engineering prior confidence table. The decay change coefficient is determined, and the value of the decay change coefficient is in the range of 0.5 to 0.9; in step S523, if the calculated node confidence value is lower than the survival threshold of 0.4, then the inferior nodes with low confidence are removed, the circuit breaker state is lifted, a globally self-consistent adaptive knowledge graph structure is constructed, and the fused self-consistent association rule instruction is output to the control terminal.

[0024] Preferably, the process of converting the initial tuple into a conditional logic tuple in step S2 includes the following sub-steps: Step S21, calling the dependency parsing operator to decouple the tree topology structure of the original sentence containing the initial tuple; Step S22, locating and extracting prepositional phrases and conditional adverbial clauses that modify related actions to define topological boundary constraint metadata; Step S23, retrieving the preset discrete ladder threshold mapping table in the storage medium, comparing the quantitative characters in the topological boundary constraint metadata, fuzzifying the continuous physical parameters into unique discrete attribute intervals, and appending and binding the discrete attribute intervals to the relational edge modifier structure of the subject-object pair.

[0025] Preferably, the process of calculating the overlap of attribute intervals in step S4 includes the following sub-steps: Step S41, obtaining the upper and lower boundaries of each unidimensional attribute interval corresponding to the first attribute interval label and the second attribute interval label in the multidimensional environmental factor space; Step S42, for each unidimensional attribute interval, comparing the lower boundary of the first attribute interval label with the upper boundary of the second attribute interval label, and comparing the upper boundary of the first attribute interval label with the lower boundary of the second attribute interval label; Step S43, if in at least one unidimensional attribute interval, the lower boundary is greater than the upper boundary, or the upper boundary is less than the lower boundary, then it is determined that the first attribute interval label and the second attribute interval label do not overlap in the multidimensional environmental factor space, and the overlap of attribute intervals is determined to be 0.

[0026] Preferably, the environmental factors in the multidimensional environmental factor space include sunshine hours, average temperature, soil moisture, diurnal temperature range, and effective accumulated temperature during the rice growth and development stage; the single-dimensional attribute interval corresponds to the numerical interval of a single environmental factor; the first attribute interval label and the second attribute interval label are generated by projecting the multidimensional environmental vector gathered by the front-end sensor onto the core physical field scale composed of temperature, humidity, and light according to the preset hedging correlation matrix.

[0027] Preferably, a time-series topology evolution dynamic compensation process is also introduced in the graph database topology network, which includes the following sub-steps after step S5: Step S53, using the historical convergence rate of the graph in the graph database topology network as the input variable, and performing weighted processing through a preset time window decay operator to calculate the topology evolution deviation for the current time period; Step S54, when the topology evolution deviation exceeds the preset safety truncation threshold, the historical constant correction rule is automatically triggered to perform step-by-step fine-tuning of the node connection weights in the graph database topology network so that the subsequent reconstructed parallel branches adaptively fit the current resting state.

[0028] Preferably, the method for determining the activation control slot threshold of the two feedforward branch paths in step S5 includes retrieving the initial environmental attributes of historically existing topological nodes in the application scenario data source from the graph database topological network, reading the historical environmental factor range corresponding to the initial environmental attributes, and using the historical environmental factor range as the reference benchmark for activation control slots, and establishing a bidirectional dependency association constraint relationship in situ in the network topology.

[0029] Preferably, the causal reasoning constraint determination after the reconstruction of the feedforward branch path in step S5 includes the following sub-steps: Step S55, when the graph database topology network receives subsequent new input data, extract the real-time environment attribute parameters carried in the new input data; Step S56, perform lossless matching between the real-time environment attribute parameters and the first attribute interval label and the second attribute interval label in the activation control slot of the two feedforward branch paths; Step S57, if the real-time environment attribute parameters fall within the attribute interval of one of the feedforward branch paths, activate the corresponding feedforward branch path to perform causal reasoning, and lock the reasoning state of the other feedforward branch path to maintain the logical self-consistency of the global knowledge graph.

[0030] Preferably, after step S523, which removes the inferior nodes with low confidence and retains the superior branches, the following steps are included: Step S524, based on the topology feedback results of the successful merging, the confidence scores of each association rule in the engineering prior confidence table are adjusted by gain or updated by attenuation, so that the engineering prior confidence table adaptively tracks the noise distribution changes of multi-source heterogeneous data sources.

[0031] Example 1: When the rice trait association analysis system processes large-scale, cross-regional crop development observation heterogeneous text data, the data input unit acquires multi-source heterogeneous associated text sequences. Using a pre-defined part-of-speech tagging operator, it identifies entity words and associated predicates in the text sequences, extracting trait association data related to genes, phenotypes, and environmental factors. This allows the system to construct an initial tuple containing subjects, objects, and associated actions within the graph database topology. Under the condition of multi-source literature knowledge fusion covering rice generation breeding experiments in different climate zones, due to the complex interference of external environmental parameters, the same controlling gene may exhibit different traits in different growth environments or developmental stages. The regulation of rice phenotypic traits often exhibits a pattern of opposite polarities, leading to logical opposition in the literal meaning of relational edges extracted from heterogeneous multi-source text sequences that point to the same subject-object pair. If traditional knowledge fusion frameworks lack precondition constraint identification methods and classify all relative causal conclusions into absolute truth assumptions through dimensionality reduction merging, the system will inevitably choose to hard-remove phenotypic divergences caused by heterogeneity of experimental conditions as input noise, resulting in the permanent loss of a large number of valuable condition-specific low-frequency key causal mechanisms, and even causing semantic drift and topological deadlock during the merging of graph database topology networks.

[0032] To address the logical conflicts in the fusion of multi-source heterogeneous text knowledge, the system initiates data access based on part-of-speech tagging operators. The relation extraction unit identifies the subject containing specific regulatory genes in rice, the object representing specific growth stage phenotypes, and the associated action predicates expressing regulatory effects, constructing an initial tuple. This initial tuple is then transmitted to the logic transformation unit. The logic transformation unit calls the dependency parsing operator to decouple the tree-like topology of the original sentence containing the initial tuple, forcibly locating and refining the prepositional phrases and conditional adverbials that modify the associated action predicates. This extracts the environmental and temporal constraint variables implicitly dependent on the associated action predicates as topological boundary constraint metadata. The environmental and temporal constraint variables are retrieved from a preset discrete ladder threshold mapping table in the storage medium, comparing quantitative characters in the text feature fragments to fuzzify continuous physical parameters into unique discrete attribute intervals. These discrete attribute intervals are then appended as attribute vectors to the relational edge modifier structure of the subject-object pair, transforming the initial tuple into a conditional logic tuple carrying a clear monotonic boundary.

[0033] The topology reorganization unit receives conditional logic tuples and, based on the discrete attribute interval boundaries defined by the topological phase boundary constraint metadata, performs spatial discretization on the conditional logic tuples within the virtual topological space, converting them into mutually independent discrete topological logic nodes carrying attribute interval labels. Based on the affirmative or negative semantics of the predicate, each discrete topological logic node is configured with exclusive positive and negative polarity state variables, thus translating complex environmental conflicts into state machine nodes divided by discrete intervals. When the conflict arbitration unit detects multiple discrete topological logic nodes pointing to the same subject-object pair with opposite polarities, it selects the opposing first and second discrete topological logic nodes. By obtaining the upper and lower boundaries of each single-dimensional attribute interval corresponding to the first and second attribute interval labels in the multi-dimensional environmental factor space, for each single-dimensional attribute interval involving sunshine hours, average temperature, soil moisture, diurnal temperature range, and effective accumulated temperature during the rice growth and development period, it compares the lower boundary of the first attribute interval label. The upper boundary of the second attribute interval label is compared with the upper boundary of the first attribute interval label and the lower boundary of the second attribute interval label. If the condition that the lower boundary is greater than the upper boundary or the upper boundary is less than the lower boundary is met in at least one unidimensional attribute interval, then the attribute interval overlap of the two discrete attribute intervals on the spatial axis or temporal axis is determined to be 0. Conversely, if the lower boundary of the first attribute interval label is less than or equal to the upper boundary of the second attribute interval label in all unidimensional attribute intervals, and the upper boundary of the first attribute interval label is greater than or equal to the lower boundary of the second attribute interval label, then the attribute interval overlap of the two is determined to be greater than 0. In this case, the overlap in the multidimensional environmental factor space is specifically quantified by calculating the product of the intersection lengths of each unidimensional attribute interval and dividing it by the product of the union lengths of each unidimensional attribute interval. The resulting overlap value ranges from 0 to 1, thus providing accurate digital input for determining the severity of overlap of attribute intervals in subsequent branch processing.

[0034] When the discrete event condition of attribute interval overlap equals 0 is met, the conflict arbitration unit determines that the logical conflict between opposing nodes is a nonlinear topological divergence caused by heterogeneity of operating conditions rather than a data authenticity conflict. The system retains the first and second discrete topological logical nodes in situ, intercepts hard removal actions based on confidence, and the scheduling unit splits two parallel feedforward branch paths in the graph database topology network. The labels of the first and second attribute intervals are written into the activation control slots of the two feedforward branch paths, respectively, so that the mutually exclusive polarity relationship edges maintain a symbiotic state under different environmental attribute boundaries. Thus, a bidirectional dependent collaborative closed loop is established in situ in the network topology structure, making the activation gating threshold of the newly split feedforward branch path highly dependent on the initial environmental attributes accumulated by the historical topological nodes under the application scenario data source. In turn, the reconstructed parallel split topology path rigidly constrains the global adaptive knowledge graph in the process of connecting to the network. Furthermore, during the operation of the graph database topology network, the system introduces a time-series topology evolution dynamic compensation process to address the inference state evolution boundary when receiving new input data. The aforementioned time window decay operator weights the historical convergence rate using a preset decay formula. Specifically, the topology evolution deviation of the current time period is equal to the historical convergence rate multiplied by the negative exponent of the natural constant, where the exponent is the difference between the current timestamp and the historical reference timestamp divided by the preset time window constant. When the calculated topology evolution deviation exceeds the preset safety truncation threshold of 0.15, the historical constant correction rule is automatically triggered. This rule fine-tunes the node connection weights in the graph database topology network using a step-by-step algorithm, with each fine-tuning increment fixed at 0.02, until the fine-tuned topology evolution deviation falls back below the safety truncation threshold, thereby enabling the subsequent reconstructed parallel branches to adaptively fit the current resting state.

[0035] To address the sudden situation where attribute intervals of multi-source heterogeneous data cannot be eliminated due to interference from non-ideal environments or errors in the original text input, the system constructs an adaptive defense based on abnormal overlap circuit breaking. When the conflict arbitration unit determines that the overlap degree of the discrete attribute interval labels corresponding to the first discrete topological logical node and the second discrete topological logical node is greater than 0 and their semantic correlation reaches a preset threshold of 0.91, the scheduling unit determines that the system has triggered a logical conflict and intercepts the graph merging thread in the graph database topology network, suspends the conflicting routing branch, retrieves the preset engineering prior confidence table from the storage medium, compares the source data source weights of the suspended conflicting routing branches, and calculates the attenuated node confidence value by multiplying the original confidence value read from the engineering prior confidence table with a determined attenuation change coefficient between 0.5 and 0.9. When the node confidence value is lower than the survival threshold of 0.4, the low-confidence node is truncated and removed to maintain the unidirectional flow of the dominant branch, release the circuit breaker state, and output the fused self-consistent association rule instruction to the control terminal.

[0036] The operation of the graph database topology network converges to the deep information mining of the syntactic dependency structure within a single heterogeneous text source. It abandons the traditional approach of searching for unique, exclusive static truth values ​​in the global knowledge space, and instead translates the logical polarity reversal caused by multidimensional environmental variables into a controllable discrete conditional control flow. It uses discrete events generated by interval intersection verification to drive parallel branch topology reconstruction, and constructs a collaborative mechanism between business object constraints and graph topology routing with extremely low system entropy increase and computational overhead. Thus, without relying on external large-scale prior biological rule bases or massive parameter deep learning models, it blocks semantic drift, topological structure collapse and system logic deadlock caused by the heterogeneity of experimental conditions in multi-source heterogeneous data fusion, and provides downstream decision control terminals with fully self-consistent association rule guarantees with causal white-box properties.

[0037] Example 2: The current rice trait association analysis system is built on a text computation unit for multi-source crop phenotypic and gene-environment effect observations. The data acquisition unit retrieves a text data stream from an open-source academic corpus of agricultural crop trait associations, covering a 30-year span and containing 50,000 rice breeding experiment documents. To simulate noise interference induced by typesetting differences and sentence variations during the multi-source document text conversion process, syntactic structure perturbation noise with an edit distance variation rate of up to 15% is actively superimposed at the input of the data processing unit to test the system's stability in extracting implicit premise constraint logic. In specific operation, the generation and superposition process of this syntactic structure perturbation noise is as follows: the system uses a random number generator to locate the character position of the text sequence in the original academic corpus, and randomly inserts, deletes, or replaces entity words or predicate characters at that position according to a specified ratio of 15%, so that the edit distance increment between the processed text sequence and the original text sequence reaches the original length. 15%, thus artificially constructing discrete grammatically distorted corpora, and continuously inputting this noisy corpus into the data processing unit. The data transformation unit involves key control parameters when controlling the syntactic analysis cache depth. The parameters are manifested as the size of the dependency parsing sliding window. The parameters are directly limited by the average sentence length and the number of grammatical nesting levels of the text to be processed. The setting of the parameters balances the integrity of the multidimensional text context causal chain parsing and the processor memory addressing load. Its control rule is that when the semantic density of polysyllabic technical terms in the text increases and the logical level of compound clauses deepens, in order to prevent the long-distance dependency association chain across clause boundaries from breaking, the sliding window size approaches the upper limit of the working interval and is specifically set to 8 consecutive sentence groups. When the sentence structure tends to be simplified, it converges to the lower limit set to 3 consecutive sentence groups. For the current complex agricultural science and technology literature, the data transformation unit determines the sliding window size to be 5 consecutive sentence groups according to the above control rules.

[0038] The multi-source heterogeneous big data mining computing platform moves the aforementioned original sequence carrying structural perturbation noise into dynamic random access memory. The main control processor calls a preset part-of-speech tagging operator to perform entity word segmentation on a text fragment containing the expression of a target gene controlling rice disease resistance traits. From this segment, the main words are extracted as target gene identifiers, and the object words are extracted as flag leaf length and action predicates for expression regulation relationships, constructing an initial tuple. The logic transformation unit calls the dependency parsing operator according to the determined sliding window size of 5 consecutive sentence groups, adaptively decomposes the tree-like dependency network of the long sentence containing the initial tuple, purifies the prepositional phrases modifying the action predicates, and separates the occurrence of the target gene in the literature. The prerequisite operating conditions for the effect; in the direct test records of the experimental group, the first input literature recorded that the disease resistance gene inhibited the elongation of the sword leaf by 12.4% under the operating conditions of 13.4h sunshine hours and 35.2℃ daily average temperature, while the second input literature recorded that the same disease resistance gene promoted the same object word by 8.6% under the operating conditions of 10.2h sunshine hours and 22.1℃ daily average temperature; the logic transformation unit, according to the built-in discrete ladder threshold mapping table, converted the above continuous physical measurement values ​​into pure text attribute intervals, and established a first attribute interval label covering 32.5℃ to 38.2℃ and a label covering 18.3℃ to 24℃.The second attribute interval label for 6℃, specifically, is represented by the aforementioned preset hedging correlation matrix as a 3-row, 5-column array of constant coefficients. The 5 columns correspond to the five single-dimensional physical features in the multidimensional environmental vector collected by the front-end sensors: sunshine duration, average temperature, soil moisture, diurnal temperature range, and effective accumulated temperature. The 3 rows correspond to the core field scaling coefficients for temperature, humidity, and illumination. When projecting the multidimensional environmental vector onto the core physical field scale, the transformation unit multiplies the five values ​​from the collected multidimensional environmental vector with the corresponding constant coefficients in the feature rows of the hedging correlation matrix and sums them. This linear superposition yields the corresponding core physical field scaling values ​​for temperature, humidity, and illumination. The resulting core physical field scaling values ​​are then compared with the scaling interval boundaries in the discrete step threshold mapping table to generate the aforementioned... The system uses first and second attribute interval labels, along with the negative and positive semantics of the predicate, to adaptively configure negative state variable feature codes 001 and positive state variable feature codes 100 for the two independent discrete topological logical nodes produced by the preceding path. In practice, the aforementioned state variable feature codes employ a 3-bit binary encoding system, where the highest bit represents the positive control polarity, the lowest bit represents the negative control polarity, and the middle bit is reserved. When the feature code is 100, it indicates that the node possesses exclusive positive promoting state variable features, while when the feature code is 001, it indicates that the node possesses negative inhibiting state variable features. During graph merging or conflict detection in the graph database topological network, the system performs a bitwise AND operation on the feature codes of the two nodes. If the result is 000, it immediately determines that the polarities of their state variable features are opposite and that they are mutually exclusive nodes.

[0039] To compare the performance of the technical solutions in dealing with the aforementioned logical polarity reversal conflict, the computing unit simultaneously introduced a control group that used a static association rule mining algorithm but lacked a condition constraint extraction mechanism, and established a multi-dimensional effect test comparison with the experimental group using the technical solution of this application. In the conflict arbitration unit, the system extracted the two discrete topological logical nodes pointing to the same gene and the same trait but with completely opposite regulatory polarities. The spatial overlap between the first attribute interval label and the second attribute interval label was calculated by the interval intersection check operator. In the pure text quantization deduction process, the system compared the difference between the lower limit value of the first attribute interval label (32.5℃) and the upper limit value of the second attribute interval label (24.6℃), and determined that the overlap of the attribute intervals was equal to 0. Under the processing path of the control group, because the system lacked the function of identifying hidden environmental boundary constraints, it directly judged these two mutually exclusive relation edges in the global knowledge network as pseudo-true value oppositions caused by data source input noise, resulting in a severe overload state where the measured semantic association conflict index reached 0.875, and the first literature containing the condition mechanism was forcibly removed. The data caused the graph thread to lock up and the pause time reached 156.4ms; under the control flow guidance of the experimental group, the discrete event judgment result with an attribute interval overlap of 0 directly served as the trigger control point of the scheduling unit. The system retained two mutually exclusive logical nodes in situ, and the scheduling unit split two parallel feedforward branch paths in the graph database topology network to absorb different attribute interval labels in situ. Furthermore, in the fault injection experiment to test the stability boundary of the test system, when the environmental parameters of the two papers were deliberately modified to both cover 22.0 When the system's attribute interval overlap is greater than 0 and the measured semantic relevance parameter reaches 0.914, this data feature activates the abnormal overlap circuit breaker mechanism. The main control processor intercepts the graph merging thread to prevent the spread of erroneous data and automatically multiplies the original confidence level of 0.75 in the engineering prior confidence level table with the preset attenuation change coefficient of 0.55 to obtain a node confidence level measured to an accuracy of 0.412. Since this value is higher than the lower limit of the survival threshold of 0.400, the system suspends the current routing branch and releases the circuit breaker.

[0040] The final measurement data from the aforementioned multidimensional comparative experiment show that the stability index of text knowledge fusion output by the experimental group increased from 62.3% in the control group to 96.8%, the cross-source logical semantic drift rate decreased from 34.2% in the control group to 1.5%, and the system maintained a stable transmission efficiency of 4200 rules per second in the parallel graph network topology routing distribution throughput under multi-threaded concurrent state. The above data proves that the method of this application can transform the logical opposition at the text semantic level into topological parallel branches controlled by phase boundary constraints. Without relying on a large external prior biological rule base, it can block semantic drift, topological structure collapse and system logic deadlock caused by the heterogeneity of experimental conditions in multi-source heterogeneous data fusion, and provide downstream decision control terminals with fully self-consistent association rules with clear causality.

[0041] Example 3: This example combines Figures 1 to 2 This paper describes a logical extraction method in rice trait association analysis, such as... Figure 1 As shown, the operation flow of this method includes step S1, identifying entity words and related predicates from multi-source heterogeneous associated text sequences to construct an initial tuple containing subjects, objects, and associated actions; step S2, analyzing the dependency syntax of the initial tuple, extracting environmental constraint variables and temporal constraint variables as topological boundary constraint metadata, thereby converting the initial tuple into a conditional logic tuple; step S3, segmenting the conditional logic tuple according to the topological boundary constraint metadata, converting it into multiple discrete topological logic nodes carrying attribute interval labels and having positive and negative polarity state variable characteristics; and then proceeding to step S4. When there are multiple discrete topological logical nodes pointing to the same subject-object pair and with opposite polarities, select the opposing first and second discrete topological logical nodes, and calculate the attribute interval overlap between the first and second attribute interval labels of the two nodes. Finally, proceed to step S5. If the attribute interval overlap is equal to 0, retain the first and second discrete topological logical nodes in place, and split parallel feedforward branch paths in the graph database topological network. Write the first and second attribute interval labels into the activation control slots respectively, so that the mutually exclusive polarity relation edges can coexist under different environmental attribute boundaries.

[0042] like Figure 2As shown, the complete data flow and logical architecture begin with the accumulation of heterogeneous text data from multi-source heterogeneous related text sequences and cross-regional multi-generation crop phenotypic observations. This data is input into an initial tuple construction structure to identify entity words and related predicates, and outputs an initial tuple containing the subject, object, and related actions. This initial tuple is then passed to a dependency parsing and decoupling structure for dependency parsing, extracting environmental and temporal constraint variables to define topological boundary constraint metadata. The extracted topological boundary constraint metadata is processed in two ways. One way is input into a discrete ladder threshold mapping table comparison structure, comparing the constraint metadata quantitative characters to fuzzify and convert continuous physical parameters and additionally bind them to the relational edge modifier structure. The other way is input together with the initial tuple into a conditional logic tuple transformation structure, converting the initial tuple into a conditional logic tuple carrying attribute interval labels. The conditional logic tuples are further fed into the discrete topology logic node transformation structure for segmentation and configuration of exclusive positive and negative polarity state variables, thereby converting them into discrete topology logic nodes. These nodes then enter the attribute interval overlap calculation structure to obtain the multidimensional environmental factor space of opposing nodes and compare it with the upper and lower limits of the single-dimensional attribute interval. Based on the calculated attribute interval overlap result, they are differentiated into two independent paths. When the attribute interval overlap is equal to zero, it is input into the parallel feedforward branch path splitting structure, where opposing logic nodes are retained in situ and written into the feedforward branch path activation control slot, allowing the polarity mutually exclusive edges to coexist under the environmental boundary. When the attribute interval overlap is greater than zero, it is input into the abnormal overlap circuit breaking branch structure. By calculating the semantic relevance, when a preset threshold is reached, the graph merging thread is intercepted and the conflict route is suspended. Finally, the node decay is determined by the engineering prior confidence table.

[0043] Example 4: In the extreme working condition calibration environment of the multi-source rice trait association analysis text knowledge fusion system, the computing platform is under extreme boundary conditions with a high proportion of syntactic distortion and deep coupling of multi-source literature noise. At this time, the input heterogeneous text sequence contains up to 25% syntactic misalignment and logical semantic inversion rate, which causes the conventional dependency parsing operator to generate a large number of pseudo-conflict nodes when extracting trait association data about gene phenotype and environmental factors. If the system lacks a closed calibration method for the core logical arbitration parameters and only relies on static empirical values, when dealing with implicit negative event record injection or facing high-density contradictory tuples, the semantic drift of the topology network will be caused by the threshold setting being too wide, or the graph merging thread will be erroneously circuit-broken and the branch route deadlock will be caused by the threshold setting being too narrow, which will lead to the entire data transformation and topology recombination process coming to a standstill.

[0044] To establish the physical action boundaries of the preset semantic relevance threshold and attenuation change coefficient used to block logical conflicts in the conflict arbitration unit, the system defines specific specifications for the input end through a control program. The data structure of the object is limited to an array of discrete topological logical nodes containing 1000 mutually exclusive features. The computing platform of the enabling environment uses a double-precision floating-point processor with a data throughput bandwidth of no less than 10GB per second. The calibration thread controls the conflict arbitration unit to gradually increase the preset semantic relevance threshold from the lower limit of 0.70 to the upper limit of 0.95 in increments of 0.05, while simultaneously adjusting the attenuation change coefficient from the lower limit of 0.50. The system increments by a step gradient of 0.05 towards an upper limit of 0.90. At each discrete parameter combination node, the system sequentially injects the original test sequence containing known interference noise into the relation extraction unit. By comparing the attribute interval overlap between the first and second discrete topological logical nodes, under the trigger condition that the overlap is greater than 0 and the semantic relevance reaches the currently set step value, the main control processor counts the frequency of intercepting and suspending the graph merging thread within a 60-second time window and the number of dominant branch nodes ultimately removed from the graph database topology network. Data analysis results show that when the preset semantic relevance threshold is set to 0.91 and the attenuation change coefficient is set... When the value is 0.55, the recognition rate of opposing nodes reaches 98.6%, and the branch suspension recovery time is 12.4 milliseconds. However, once the parameter exceeds the upper limit or falls below the lower limit, it will cause the node survival rate to drop below 60% or induce continuous deadlock in the graph topology. Therefore, relying on this data change trend, an engineering judgment benchmark with a preset semantic relevance threshold of 0.91 and a decay change coefficient of 0.55 is locked. Specifically, the aforementioned semantic relevance calculation model is completed by extracting the text vectors of two sets of discrete topological logical nodes and calculating the cosine similarity score. The selection of the aforementioned threshold and coefficient is based on the fact that when the semantic relevance reaches 0.91, it indicates that the two If two opposing nodes have a high degree of semantic overlap in the text context, and the overlap of attribute intervals is greater than 0, it is highly likely that they belong to the same experimental conditions and are data entry errors rather than environmental discrepancies. In this case, the circuit breaker is activated, and the attenuation change coefficient is set between 0.5 and 0.9 to ensure that low-confidence inferior nodes are appropriately penalized. After the parameters in this range are multiplied and attenuated, if the obtained confidence is lower than the survival threshold of 0.4, it means that the node contains too much noise and removing it can ensure global consistency. If it is higher than 0.4, it is suspended for subsequent verification, thus balancing the noise filtering effect and the retention of key mechanisms in engineering.

[0045] When the multi-source rice trait association analysis text knowledge fusion system is running stably under the adjusted parameter boundaries, the conflict arbitration unit receives the conditional logic tuples output by the logic transformation unit. Based on the phase boundary overlap, it determines the parallel branching topology path generated by the logic and reconstruction. The system assigns the polar mutually exclusive relation edges to independent discrete attribute intervals for coexistence. Even if subsequent data sources continuously inject implicit negative event records containing high noise, the conflict arbitration unit intercepts the graph merging thread and calculates the decay product value based on the locked preset semantic relevance threshold of 0.91 and survival threshold of 0.4. Due to the phenotypic trait divergence caused by the heterogeneity of experimental conditions, adaptive parallel routing splits occur on the spatial and temporal axes. The data fusion error rate in the graph database topology network steadily decreases from the original 35.6% to 1.2%, controlling the semantic drift phenomenon across information sources. It delivers fully self-consistent causal association rules to the downstream breeding decision control terminal, showing the causal self-consistent topological skeleton of the nonlinear interaction relationship between rice control genes, growth period phenotypes, and environmental factors.

[0046] Example 5: When the system faces the initial deployment of literature source data for new crop varieties, the main control processor initiates a pre-deployment calibration procedure before starting the data transformation process. It retrieves the benchmark sample text sequence from the storage array, calls the dependency parsing operator to extract the numerical distribution characteristics containing environmental and temporal constraint variables, and calculates the statistical variance and numerical boundary envelope of quantitative characters in the sample text fragments for physical parameters such as sunshine hours, average temperature, and effective accumulated temperature. The continuous physical parameters are divided into non-overlapping discrete gradient intervals by calculating the statistical variance and numerical boundary envelope of quantitative characters in the sample text fragments. The data of each interval is written as the basic data unit into the corresponding storage slot of the offline step threshold mapping table to establish the monotonicity mapping criterion when converting the initial tuple to the conditional logic tuple.

[0047] After updating the offline ladder threshold mapping table and attaching the conflict arbitration unit, the computing platform inputs discrete topological logic nodes with attribute interval labels into the graph database topological network. The conflict arbitration unit performs interval intersection verification on the first and second discrete topological logic nodes with mutually exclusive control polarities according to the discrete gradient interval boundaries filled by the pre-calibration procedure, and calculates the attribute interval overlap degree of the two. When it is determined that the overlap degree of the sunshine hours and average temperature of the two on the spatial axis or temporal axis is equal to 0, the scheduling unit directly constructs parallel feedforward branch paths in the graph database topological network, splits and co-generates the relationship edges with opposite polarities, controls the semantic drift phenomenon across information sources, and makes the global reasoning state evolution boundary controlled by the self-consistent topological skeleton network structure.

[0048] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A logical extraction method for rice trait association analysis, characterized in that, include: Step S1: Identify entity words and related predicates from multi-source heterogeneous associated text sequences, and construct an initial tuple containing subject, object and associated action; Step S2: Analyze the dependency syntax of the initial tuple, extract the environmental constraint variables and temporal constraint variables as topological phase boundary constraint metadata, and convert the initial tuple into a conditional logic tuple. Step S3: Based on the topological phase boundary constraint metadata, the conditional logic tuples are segmented and converted into multiple discrete topological logic nodes carrying attribute interval labels and having positive and negative polarity state quantity characteristics. Step S4: When there are multiple discrete topological logic nodes pointing to the same subject-object pair and with opposite polarities, select the opposing first discrete topological logic node and the second discrete topological logic node, and calculate the overlap degree of the attribute intervals between the first attribute interval label and the second attribute interval label of the two nodes. Step S5: If the overlap of attribute intervals is 0, retain the first discrete topological logical node and the second discrete topological logical node in place, and split two parallel feedforward branch paths in the graph database topological network. Write the first attribute interval label and the second attribute interval label into the activation control slots of the two feedforward branch paths respectively, so that the polarity mutually exclusive relational edges can coexist under different environmental attribute boundaries.

2. The logical extraction method in rice trait association analysis according to claim 1, characterized in that, If the overlap of attribute intervals is greater than 0 in step S5, abnormal overlap circuit breaking is performed. Abnormal overlap circuit breaking includes the following sub-steps: Step S51, calculate the semantic relevance between the first discrete topology logical node and the second discrete topology logical node; Step S52, if the semantic relevance reaches the preset threshold of 0.91, intercept the graph merging thread in the graph database topology network and suspend the conflicting routing branches.

3. The logical extraction method in rice trait association analysis according to claim 2, characterized in that, After step S52, the following sub-steps are included: Step S521, retrieve the preset engineering prior confidence table in the storage medium and compare the source data source weights of the suspended conflicting routing branches. Step S522: The confidence decay formula is used to determine the decay of weak nodes with low confidence. The confidence decay formula is expressed as follows: in, This represents the node confidence score after decay. These are the original confidence values ​​read from the engineering prior confidence table. The decay change coefficient is determined, and the value of the decay change coefficient is in the range of 0.5 to 0.9; in step S523, if the calculated node confidence value is lower than the survival threshold of 0.4, then the inferior nodes with low confidence are removed, the circuit breaker state is lifted, a globally self-consistent adaptive knowledge graph structure is constructed, and the fused self-consistent association rule instruction is output to the control terminal.

4. The logical extraction method in rice trait association analysis according to claim 1, characterized in that, The process of converting the initial tuple into a conditional logic tuple in step S2 includes the following sub-steps: Step S21, calling the dependency parsing operator to decouple the tree topology of the original sentence containing the initial tuple; Step S22, locating and extracting prepositional phrases and conditional adverbial clauses that modify related actions to define topological boundary constraint metadata; Step S23, retrieving the preset discrete ladder threshold mapping table in the storage medium, comparing the quantitative characters in the topological boundary constraint metadata, fuzzifying the continuous physical parameters into unique discrete attribute intervals, and appending and binding the discrete attribute intervals to the relational edge modifier structure of the subject-object pair.

5. The logical extraction method in rice trait association analysis according to claim 1, characterized in that, The process of calculating the overlap of attribute intervals in step S4 includes the following sub-steps: Step S41, obtaining the upper and lower boundaries of each unidimensional attribute interval corresponding to the first attribute interval label and the second attribute interval label in the multidimensional environmental factor space; Step S42, for each unidimensional attribute interval, comparing the lower boundary of the first attribute interval label with the upper boundary of the second attribute interval label, and comparing the upper boundary of the first attribute interval label with the lower boundary of the second attribute interval label; Step S43, if in at least one unidimensional attribute interval, the lower boundary is greater than the upper boundary, or the upper boundary is less than the lower boundary, then it is determined that the first attribute interval label and the second attribute interval label do not overlap in the multidimensional environmental factor space, and the overlap of attribute intervals is determined to be 0.

6. The logical extraction method in rice trait association analysis according to claim 5, characterized in that, The environmental factors in the multidimensional environmental factor space include sunshine hours, average temperature, soil moisture, diurnal temperature range and effective accumulated temperature during the rice growth and development stage; the single-dimensional attribute interval corresponds to the numerical interval of a single environmental factor; the first attribute interval label and the second attribute interval label are generated by projecting the multidimensional environmental vector gathered by the front-end sensor onto the core physical field scale composed of temperature, humidity and light according to the preset hedging correlation matrix.

7. The logical extraction method in rice trait association analysis according to claim 1, characterized in that, In the graph database topology network, a time-series topology evolution dynamic compensation process is also introduced, which includes the following sub-steps after step S5: Step S53, using the historical convergence rate of the graph in the graph database topology network as the input variable, and performing weighted processing through a preset time window decay operator, the topology evolution deviation of the current time period is calculated; Step S54, when the topology evolution deviation exceeds the preset safety truncation threshold, the historical constant correction rule is automatically triggered to perform step-by-step fine-tuning of the node connection weights in the graph database topology network, so that the subsequent reconstructed parallel branches can adaptively fit the current resting state.

8. The logical extraction method in rice trait association analysis according to claim 1, characterized in that, The method for determining the activation control slot thresholds of the two feedforward branch paths in step S5 includes retrieving the initial environmental attributes of historically existing topological nodes in the application scenario data source from the graph database topological network, reading the historical environmental factor range corresponding to the initial environmental attributes, and using the historical environmental factor range as the reference benchmark for activation control slots to establish bidirectional dependency association constraints in situ in the network topology.

9. The logical extraction method in rice trait association analysis according to claim 1, characterized in that, The causal reasoning constraint determination after the reconstruction of the feedforward branch path in step S5 includes the following sub-steps: Step S55, when the graph database topology network receives subsequent new input data, extract the real-time environment attribute parameters carried in the new input data; Step S56, perform lossless matching between the real-time environment attribute parameters and the first attribute interval label and the second attribute interval label in the activation control slot of the two feedforward branch paths; Step S57, if the real-time environment attribute parameters fall within the attribute interval of one of the feedforward branch paths, activate the corresponding feedforward branch path to perform causal reasoning, and lock the reasoning state of the other feedforward branch path to maintain the logical self-consistency of the global knowledge graph.

10. The logical extraction method in rice trait association analysis according to claim 3, characterized in that, After step S523 removes the inferior nodes with low confidence and retains the superior branches, the following steps are included: Step S524, based on the topology feedback results of the successful merging, the confidence scores of each association rule in the engineering prior confidence table are adjusted by gain or updated by attenuation, so that the engineering prior confidence table adaptively tracks the noise distribution changes of multi-source heterogeneous data sources.

Images