Intelligent exploratory data mining system

Abstract

The invention relates to the technical field of data mining, and discloses a feasibility research data intelligent mining system which comprises a heterogeneous data acquisition module, a semantic association analysis module, a decision map generation module, a time sequence feature correction module and a knowledge distillation optimization module. The heterogeneous data acquisition module captures data features through a multi-source data sensing node and a dynamic dimension fusion network and constructs a multi-layer topology; the semantic association analysis module analyzes semantic association by using a concept topology modeling unit, a knowledge vector clustering unit and a multi-mode switching link; the decision graph generation module generates a core decision reference framework based on strategy optimization nodes and a rule inference engine; the time sequence feature correction module performs time sequence correction and noise compensation on the semantic association; and the knowledge distillation optimization module detects the deviation through the entity relationship evaluation network and feeds back the optimization decision framework. The system realizes multi-source data intelligent acquisition, semantic dynamic analysis and decision graph adaptive generation, and improves the accuracy and efficiency of feasibility research data mining.

Description

Technical Field

[0001] The present invention relates to the technical field of data mining, and in particular to an intelligent feasibility data mining system. Background Art

[0002] In the modern data-driven decision-making environment, feasibility study (FSS) data mining faces complex, multi-dimensional, cross-domain challenges. Traditional data mining systems, when processing heterogeneous data, suffer from common problems such as limited data collection dimensions, insufficient semantic association parsing accuracy, and poor dynamic adaptability of decision models. These systems struggle to meet the real-time, accurate, and comprehensive requirements of FS scenarios.

[0003] From a data collection perspective, feasibility study data typically includes structured data (such as numerical statistics and tabular data) and unstructured data (such as text reports, image information, and voice recordings). The formats, semantic rules, and update frequencies of these different data types vary significantly. Traditional systems often employ fixed data collection models and are unable to dynamically perceive the meta-attribute characteristics and latent semantic labels of multi-source data. This results in incomplete data feature extraction and makes it difficult to construct a topological structure that reflects the inherent connections within the data. For example, in a financial feasibility study scenario, implicit semantic connections exist between structured financial data and unstructured industry news text. Traditional collection methods struggle to effectively capture these connections, impacting the accuracy of subsequent analysis.

[0004] When it comes to semantic association analysis, existing technologies rely heavily on static concept modeling and single clustering algorithms, lacking the ability to respond to dynamic changes in data semantics. As data volumes grow and domain knowledge is updated, the semantic associations between data exhibit temporal fluctuations and noise interference. Traditional systems are unable to correct and compensate for these dynamic changes in real time, resulting in increased concept mapping deviations and decreased reliability of knowledge vector clustering results. For example, in medical feasibility studies, the semantic associations between emerging disease terms and existing diagnostic criteria require real-time updates. Traditional analysis methods struggle to quickly adapt to these changes, potentially leading to lags or deviations in the basis for decision-making.

[0005] When generating decision maps, traditional systems typically employ fixed rule engines and policy frameworks, lacking the ability to dynamically adjust the weights of multi-source data or integrate cross-domain knowledge. When faced with complex feasibility studies, they are unable to generate a flexible decision-making benchmark framework based on the real-time state and correlation density of the data, nor are they able to dynamically optimize information extraction paths. This results in insufficient practicality and targetedness in decision maps. For example, in urban planning feasibility studies, data from multiple fields, including the economy, environment, and society, is involved. Traditional systems struggle to effectively integrate cross-domain knowledge and generate decision maps that address multi-dimensional needs, potentially leading to one-sided planning solutions.

[0006] Furthermore, existing systems lack effective closed-loop feedback mechanisms and adaptive optimization capabilities for dealing with data anomalies and noise interference. When data exhibits unusual fluctuations or noise interference, traditional systems are unable to detect and correct them promptly. This can lead to cumulative errors throughout the mining process, compromising the reliability of the final decision. For example, in industrial production feasibility studies, if sudden noise in sensor data cannot be compensated for promptly, it can lead to misjudgments of the production process, further impacting production planning. Summary of the Invention

[0007] The purpose of the present invention is to provide an intelligent feasibility data mining system to solve the problems raised in the above background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: an intelligent feasibility study data mining system, the system comprising: Heterogeneous data acquisition module, semantic association analysis module and decision graph generation module; The heterogeneous data acquisition module includes a multi-source data perception node and a dynamic dimension fusion network; the semantic association analysis module includes a concept topology modeling unit and a knowledge vector clustering unit; the decision map generation module includes a strategy optimization node and a rule reasoning engine; the multi-source data perception node is used to capture the meta-attribute characteristics of structured data and the potential semantic labels of unstructured data in real time; according to the data update frequency and entity association density, a multi-layer data feature topology is constructed through the dynamic dimension fusion network; the dynamic dimension fusion network is composed of M domain threshold-based fusion subnets connected in series; based on the real-time semantic response characteristics of the M fusion subnets, a core decision benchmark framework is generated through the strategy optimization node, and combined with the association constraint signal output by the knowledge vector clustering unit, the information extraction path is dynamically weighted and sorted through the rule reasoning engine.

[0009] Preferably, the system further comprises a time series feature correction module and a knowledge distillation optimization module; the time series feature correction module comprises a periodic fluctuation suppression unit and a noise interference compensation unit; According to the generated core decision benchmark framework and dynamic weight ranking results, the data semantic association range is corrected in time series through the periodic fluctuation suppression unit, and the vector parameters of the semantic association parsing module are dynamically smoothed through the noise interference compensation unit; at the same time, the concept mapping deviation rate is detected in real time through the entity relationship evaluation network constructed by the knowledge distillation optimization module, and the detection results are fed back to the strategy optimization node and rule inference engine to perform closed-loop iteration on the decision framework.

[0010] Preferably, the periodic fluctuation suppression unit is composed of feature transmission channels corresponding to fusion subnets connected in series in the dynamic dimension fusion network; the steps of constructing the multi-layer data feature topology include: Capture meta-attribute features and semantic label fluctuation signals, input them into the dynamic dimension projection model configured in each level of fusion subnet, generate data correlation compensation coefficients and store them in the corresponding fusion subnet; Based on the correlation compensation coefficients stored in fusion subnets at all levels and combined with the instantaneous change rate of entity relationships, the information extraction weights are distributed to M fusion subnets through the feature importance evaluation algorithm to form a dynamic feature matching topology.

[0011] Preferably, the fusion subnet construction conditions of the multi-source data perception node configuration include: the meta-attribute feature is within a preset domain range, the entity recognition rate corresponding to the fusion subnet is greater than a threshold, and the semantic association density is less than an allowable deviation range; The conceptual topology modeling unit is also configured with a multimodal switching link; the multimodal switching link includes a steady-state association sub-link and an abnormal blocking sub-link; the steady-state association sub-link is generated based on the matching degree between the entity relationship signal of the semantic association parsing module and the core decision reference framework; the abnormal blocking sub-link is constructed based on the association relationship between the instantaneous conflict risk of the information extraction path and the data noise event.

[0012] Preferably, the strategy optimization node is configured with a cross-domain knowledge fusion model; the cross-domain knowledge fusion model includes an attribute-semantic mapping table corresponding to the heterogeneous data acquisition module and an abnormal blocking parameter library of the semantic association analysis module; the steps of generating the core decision benchmark framework include: Based on the real-time attribute data and weight ranking results of the heterogeneous data acquisition module, the theoretical upper limit of the association range of each level of fusion subnet in the steady-state association sub-link is calculated; The upper limit of the theoretical correlation range and transient data noise are input into the cross-domain knowledge fusion model to generate a decision logic reference map, and the conflict risk of the reference map is corrected through the abnormal blocking sub-link to form a core decision benchmark framework.

[0013] Preferably, the parameter updating step of the cross-domain knowledge fusion model includes: When it is detected that the meta-attribute feature exceeds the preset domain range or the concept mapping deviation rate exceeds the threshold, the first abnormal blocking instruction is triggered, that is, the replacement knowledge unit is started and the parsing granularity of the semantic association parsing module is adjusted; If the associated range has not been restored to the allowed range after the execution of the first exception blocking instruction, the second exception blocking instruction is triggered, that is, switching to the exception blocking sub-link through the multimodal switching link, and reallocating the topological weight of the information extraction path based on the duration of the data noise event.

[0014] Preferably, the parameter updating step of the cross-domain knowledge fusion model further includes: When the entity recognition rate of a certain level of fusion subnet in the heterogeneous data acquisition module is lower than the threshold, the third abnormal blocking instruction is triggered, that is, the associated output channel of the subnet is blocked, and the corresponding data flow is transferred to other fusion subnets through the periodic fluctuation suppression unit; During the execution of the third exception blocking instruction, if other fusion subnets are detected to be overloaded, the fourth exception blocking instruction is triggered, that is, the knowledge distillation optimization module is called to downgrade the output strategy of the rule inference engine and limit the association requirements of the edge field.

[0015] Preferably, the degradation processing logic of the knowledge distillation optimization module includes: Based on the deviation rate data and knowledge vector clustering indicators output by the entity relationship evaluation network, a domain-related security level table is constructed; When the fourth abnormal blocking instruction is triggered, the priority of the associated path is dynamically downgraded according to the security level table, and the parsing frequency of the semantic association parsing module is synchronously adjusted to match the downgraded decision-making strategy.

[0016] Preferably, the step of constructing the fusion subnet further includes: Generate a domain gradient adaptation index table based on the data clustering results of meta-attribute features and the fluctuation range of historical semantic labels; According to the difference in correlation compensation coefficients of each fusion subnetwork in the index table, the semantic weighted regression algorithm is used to distribute topological weights to the feature transmission channels of the M fusion subnetworks, and redundant association links are established between the fusion subnetworks.

[0017] Preferably, the process of generating the decision logic reference graph includes: After inputting the theoretical correlation range upper limit and transient data noise into the cross-domain knowledge fusion model, the preset knowledge graph database and data anomaly interference layer are loaded simultaneously; Based on the ontology relationship strength data output by the concept topology modeling unit, the conflict threshold of the information extraction path in the reference graph is corrected, and the data sparsity parameter is superimposed in the abnormal blocking sub-link for secondary calibration.

[0018] Compared with the prior art, the present invention has the following beneficial effects: In terms of heterogeneous data collection, multi-source data perception nodes can capture the meta-attribute characteristics of structured data and the latent semantic labels of unstructured data in real time, breaking through the traditional system's limitation of collecting single data types and achieving comprehensive perception of multi-source data. The dynamic dimension fusion network is composed of M domain-threshold-based fusion subnetworks connected in series, which can construct a multi-layer data feature topology based on data update frequency and entity association density. This structural design not only adapts to the characteristic differences of different data types, but also dynamically adjusts the data feature fusion method through real-time semantic response characteristics, ensuring that the collected data can fully and accurately reflect the complex associations in real-world scenarios, providing a rich and reliable data foundation for subsequent semantic analysis and decision-making.

[0019] The semantic association parsing module achieves in-depth analysis of data semantic associations through the collaborative work of the concept topology modeling unit and the knowledge vector clustering unit. The multimodal switching link configured by the concept topology modeling unit includes a steady-state association sub-link and anomaly blocking sub-link. This dynamically switches parsing modes based on the matching degree between entity relationship signals and the core decision benchmark framework, as well as the correlation between the instantaneous conflict risk of information extraction paths and data noise events, effectively addressing dynamic changes in data semantics and abnormal interference. The association constraint signals output by the knowledge vector clustering unit, combined with the real-time semantic response characteristics of the dynamic dimension fusion network, dynamically weight-rank information extraction paths, improving the accuracy and flexibility of semantic parsing. Furthermore, the periodic fluctuation suppression unit and noise interference compensation unit in the temporal feature correction module perform time series correction on the data semantic association range and dynamically smooth vector parameters, further reducing the impact of temporal fluctuations and noise interference on semantic parsing and ensuring the stability and reliability of semantic association parsing results.

[0020] The decision graph generation module achieves intelligent generation and optimization of the decision framework through the innovative design of policy optimization nodes and rule inference engines. The cross-domain knowledge fusion model configured in the policy optimization nodes can integrate cross-domain knowledge, such as the attribute-semantic mapping table of the heterogeneous data acquisition module and the anomaly blocking parameter library of the semantic association analysis module, to generate a core decision benchmark framework. This framework calculates the upper limit of the theoretical association range of each fusion subnet in the steady-state association sub-link, and combines it with conflict risk correction for transient data noise and anomaly blocking sub-links to ensure that the decision framework can accurately reflect the real-time status and potential associations of the data. The rule inference engine optimizes the information extraction path based on dynamic weight sorting results and association constraint signals, enabling the decision graph to dynamically adjust the focus and direction of information extraction according to actual needs, thereby improving the practicality and pertinence of the decision graph.

[0021] The system's temporal feature correction module and knowledge distillation optimization module establish a comprehensive closed-loop feedback mechanism. The temporal feature correction module detects temporal fluctuations and noise interference in data semantic associations in real time and dynamically corrects and compensates for them. The knowledge distillation optimization module uses the entity relationship evaluation network to detect concept mapping deviation rates in real time and feeds the results back to the policy optimization node and rule inference engine, achieving a closed-loop iteration of the decision framework. This feedback mechanism enables the system to self-optimize and adjust based on the real-time status of the data and processing results, effectively addressing data anomalies and noise interference, and improving the system's robustness and adaptability.

[0022] During the construction of the fusion subnetwork, a domain gradient adaptation index table is generated based on the data clustering results of meta-attribute features and the fluctuation range of historical semantic labels. Furthermore, topological weights are assigned and redundant association links are established using a semantic weighted regression algorithm, further improving the efficiency and reliability of data feature fusion. This design enables the system to dynamically adjust the operating mode and weight distribution of the fusion subnetwork based on the inherent characteristics and historical patterns of the data, ensuring that effective information is retained and utilized to the greatest extent possible during the data collection and fusion process, reducing the risk of data loss and misjudgment.

[0023] The cross-domain knowledge fusion model's parameter update mechanism and hierarchical triggering strategy for exception blocking instructions enable the system to promptly and effectively handle data anomalies and parsing deviations. By initiating alternative knowledge units, adjusting parsing granularity, switching parsing links, shifting data streams, and downgrading processing, the system maintains stable operation under varying degrees of anomalies, ensuring the continuity of the data mining process and the reliability of decision-making results. This multi-level exception handling mechanism significantly improves the system's fault tolerance and anti-interference capabilities in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a diagram showing the working principle of the intelligent feasibility study data mining system described in the present invention. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0026] See also Figure 1 The present invention provides an intelligent feasibility data mining system, which includes a heterogeneous data acquisition module, a semantic association analysis module and a decision graph generation module. The specific structure and implementation method of each module are described in detail below.

[0027] The heterogeneous data acquisition module includes a multi-source data perception node and a dynamic dimension fusion network. The multi-source data perception node is used to capture the meta-attribute features of structured data and the latent semantic labels of unstructured data in real time. Specifically, for structured data, the module connects to databases and APIs through pre-set data interfaces to extract meta-attribute features such as field names, data types, and value ranges. For unstructured data, such as text, images, and audio, the module utilizes natural language processing, image recognition, and speech recognition technologies to parse the semantic information in the data and generate latent semantic labels. For example, it extracts keywords and entity names from text and object categories and color features from images.

[0028] After capturing data features, a multi-layer data feature topology is constructed using a dynamic dimensional fusion network based on data update frequency and entity association density. This dynamic dimensional fusion network consists of M domain-threshold-based fusion subnetworks connected in series, each corresponding to a different domain threshold range, for hierarchical data feature processing. The specific construction steps are as follows: First, meta-attribute features and semantic label fluctuation signals are captured and input into a dynamic dimensional projection model configured in each fusion subnetwork level. This model analyzes the input signals using a preset algorithm to generate data correlation compensation coefficients, which are then stored in the corresponding fusion subnetwork. Then, based on the correlation compensation coefficients stored in each fusion subnetwork and the instantaneous change rate of entity relationships, a feature importance evaluation algorithm is used to calculate the importance of each fusion subnetwork for information extraction. Information extraction weights are then distributed to the M fusion subnetworks, forming a dynamic feature matching topology and achieving multi-layered, dynamic fusion processing of data features.

[0029] The semantic association analysis module includes a concept topology modeling unit and a knowledge vector clustering unit. The concept topology modeling unit is used to construct a concept association topology structure between data. By analyzing the metadata attributes and semantic labels of the data, it establishes association relationships between concepts and forms a concept network. The knowledge vector clustering unit clusters the concept vectors output by the concept topology modeling unit, grouping concepts with similar semantics into one category and outputting association constraint signals for subsequent decision map generation.

[0030] The decision graph generation module includes a policy optimization node and a rule inference engine. The policy optimization node generates a core decision benchmark framework based on the real-time semantic response characteristics of M fusion subnets. The specific process is: according to the real-time attribute data and weight sorting results of the heterogeneous data acquisition module, the theoretical correlation range upper limit of each level of fusion subnet in the steady-state correlation sub-link is calculated; the theoretical correlation range upper limit and transient data noise are input into the cross-domain knowledge fusion model, which includes the attribute-semantic mapping table corresponding to the heterogeneous data acquisition module and the abnormal blocking parameter library of the semantic association analysis module. The decision logic reference graph is generated through model analysis, and the reference graph is corrected for conflict risks through the abnormal blocking sub-link to form a core decision benchmark framework. The rule inference engine combines the association constraint signal output by the knowledge vector clustering unit to dynamically weight the information extraction path to optimize the information extraction process.

[0031] Example 1: This embodiment involves the system's expansion modules and time series feature correction functionality. In addition to the heterogeneous data acquisition module, semantic association analysis module, and decision graph generation module, the system also includes a time series feature correction module and a knowledge distillation optimization module. The time series feature correction module further includes a periodic fluctuation suppression unit and a noise interference compensation unit. These modules, through a specific logical architecture and data interaction mechanism, implement time series correction for data semantic associations, noise processing, and closed-loop optimization of the decision framework.

[0032] 1. Structure and Functional Implementation of Timing Feature Correction Module The cyclical fluctuation suppression unit consists of feature transfer channels corresponding to the serially connected fusion subnetworks in the dynamic dimensional fusion network. During system operation, the M fusion subnetworks of the dynamic dimensional fusion network are serially connected in a preset order. Each fusion subnetwork corresponds to a specific domain threshold range and is used to perform hierarchical processing of data features. As the data transmission path between fusion subnetworks, the feature transfer channel not only performs data flow but also integrates a temporal feature perception mechanism. Specifically, after the multi-source data perception nodes capture meta-attribute features of structured data (such as field update timestamps and value fluctuation frequency) and latent semantic labels of unstructured data (such as implicit temporal adverbials in text and image capture time metadata), these time-related feature signals are fed into the dynamic dimensional fusion network along with the data. The cyclical fluctuation suppression unit uses the feature transfer channels to monitor the time series characteristics of the output data from each fusion subnetwork in real time, such as periodic fluctuations in the semantic relevance of the data (e.g., regular changes on a daily, weekly, or monthly basis).

[0033] For detected periodic fluctuations, the periodic fluctuation suppression unit performs the following processing: First, it obtains the feature data and timestamps output by each fusion subnetwork through the feature transfer channel. Using a time series analysis algorithm (such as the autoregressive moving average (ARMA) model), it models the fluctuation pattern of the data's semantic association range, identifying periodic components (such as sine and square waves) and non-periodic noise. Then, based on the modeling results, it generates an adjustment signal, which is fed back to the corresponding fusion subnetwork through the feature transfer channel. This signal adjusts the processing parameters within the fusion subnetwork (such as the weight coefficients of the dynamic dimensional projection model and the threshold of the feature importance assessment algorithm) to mitigate the impact of periodic fluctuations on the data semantic association analysis. For example, if the semantic association range of a certain type of data is found to fluctuate significantly between 9:00 and 11:00 AM daily, the periodic fluctuation suppression unit can send parameter adjustment instructions to the corresponding fusion subnetwork through the feature transfer channel, causing it to automatically enhance the smoothing of the fluctuating features during this period.

[0034] The noise interference compensation unit dynamically smooths the vector parameters of the semantic association parsing module. When constructing the concept association topology, the concept topology modeling unit in the semantic association parsing module converts data features into concept vectors in a high-dimensional vector space. The knowledge vector clustering unit then clusters these vectors. During this process, noise in the data (such as outliers and random interference signals) can cause errors in the concept vectors, affecting the accuracy of the clustering results. The noise interference compensation unit achieves dynamic smoothing through the following mechanism: First, the concept vectors output by the concept topology modeling unit are collected in real time. Outliers in the vectors are detected using statistical methods (such as the Z-score algorithm) to identify noisy data points. Then, for each detected noise point, a neighborhood smoothing algorithm (such as the K-nearest neighbor algorithm) is applied. Based on the distribution characteristics of the normal vectors surrounding the noise point, a corrected vector value is generated to replace the original noise vector. Furthermore, the noise interference compensation unit tracks the changing trends of the vector parameters and dynamically filters them using exponential smoothing to reduce the perturbations of high-frequency noise on the vector space structure, ensuring the stability and accuracy of the association constraint signals output by the knowledge vector clustering unit.

[0035] 2. Entity Relationship Evaluation and Feedback Mechanism of the Knowledge Distillation Optimization Module The core component of the knowledge distillation optimization module is the entity relationship evaluation network, which is responsible for real-time detection of concept mapping deviation rates. The entity relationship evaluation network is built based on a deep learning architecture. Its input is the concept association topology structure output by the concept topology modeling unit and the clustering results output by the knowledge vector clustering unit. The output is the concept mapping deviation rate. The specific implementation process is as follows: First, the concept nodes and their associated edges in the concept association topology are converted into numerical features (such as node degree and edge weight), and the clustering results are converted into label vectors (such as the cluster category to which each concept belongs). Then, these features and labels are jointly analyzed through the neural network model in the entity relationship evaluation network (such as the convolutional network GCN in the figure), and the deviation value between the mapping position of each concept in the cluster space and the ideal position is calculated, thereby obtaining the overall concept mapping deviation rate.

[0036] The detection results of the entity relationship evaluation network are transmitted to the policy optimization node and the rule inference engine through the feedback link. For the policy optimization node, when a signal is received that the concept mapping deviation rate exceeds the preset threshold, the adjustment process of the core decision benchmark framework will be triggered. Specifically, the policy optimization node will re-call the cross-domain knowledge fusion model, generate a new decision logic reference map based on the current deviation data, and correct the conflict risk through the abnormal blocking sub-link, thereby updating the core decision benchmark framework. For example, if a high concept mapping deviation rate is detected in a certain field, the policy optimization node may adjust the upper limit of the theoretical association range of the corresponding fusion sub-network in the steady-state association sub-link to adapt to the actual changes in concept associations.

[0037] For the rule-based inference engine, feedback signals from the entity relationship evaluation network are used to adjust the dynamic weight ranking of information extraction paths. Based on the deviation rate data, the rule-based inference engine identifies information extraction paths where concepts with mapping deviations exist. It then uses weight adjustment algorithms (such as priority queue algorithms) to reduce the weights of these paths while increasing the weights of other accurate paths, thereby optimizing the overall efficiency of information extraction. Furthermore, the knowledge distillation optimization module establishes a linkage mechanism with the cross-domain knowledge fusion model. When a persistently high deviation rate is detected, the parameter update process of the cross-domain knowledge fusion model is triggered (such as the exception blocking instruction mechanism described in Example 4), achieving closed-loop iterative optimization of the system's overall processing logic.

[0038] 3. Collaborative Workflow of Temporal Feature Correction and Knowledge Distillation Optimization During system operation, the Time Series Feature Correction Module and the Knowledge Distillation Optimization Module form a collaborative processing mechanism. The specific process is as follows: First, the data captured by the Heterogeneous Data Acquisition Module is processed by the Dynamic Dimension Fusion Network to generate a multi-layered data feature topology with time series characteristics. This topology structure incorporates both the spatial correlation characteristics and temporal evolution characteristics of the data. The Periodic Fluctuation Suppression Unit of the Time Series Feature Correction Module analyzes the temporal features in the topology through the feature transfer channel to suppress periodic fluctuations. The Noise Interference Compensation Unit smoothes the vector parameters of the Semantic Association Parsing Module to reduce the impact of noise.

[0039] After time-correction and noise processing, the data features enter the semantic association parsing module. The concept topology modeling unit constructs an updated concept association topology, and the knowledge vector clustering unit outputs optimized association constraint signals. The entity relationship evaluation network of the knowledge distillation optimization module monitors the mapping accuracy of the concept association topology in real time. If an abnormal deviation rate is detected, the signal is immediately fed back to the policy optimization node and the rule inference engine. Based on the feedback signal, the policy optimization node adjusts the core decision benchmark framework, and the rule inference engine adjusts the weights of the information extraction path. This simultaneously triggers the parameter update mechanism of the cross-domain knowledge fusion model, forming a closed-loop "detection-feedback-adjustment" process.

[0040] During this collaborative process, modules interact with each other via standardized data interfaces, ensuring accurate and real-time data transmission. For example, the adjustment signals sent by the periodic fluctuation suppression unit to the fusion subnet via the feature transmission channel, and the deviation rate data sent by the entity relationship evaluation network to the policy optimization node, all adhere to pre-set data formats and communication protocols. Furthermore, the system incorporates a fault-tolerant mechanism. If a module fails, other modules can continue to operate via redundant links, ensuring overall system stability and reliability.

[0041] 4. Key technical details and parameter settings The time series analysis algorithm for the periodic fluctuation suppression unit can be dynamically selected based on the data type. For structured data (such as numerical time series), models such as ARIMA and SARIMA are preferred. For unstructured data (such as temporal semantics in text), time series analysis is performed using the temporal expression recognition (TimeML) technique in natural language processing combined with the hidden Markov model (HMM). The number of feature transmission channels is consistent with the number of fusion subnetworks M, and each channel corresponds to an independent data transmission link, ensuring the accurate transmission of timing adjustment signals.

[0042] In the neighborhood smoothing algorithm of the noise interference compensation unit, the K value is dynamically adjusted based on the dimension of the concept vector and the cluster density, typically ranging from 3 to 10. The smoothing coefficient α of the exponential smoothing method is determined by an adaptive algorithm, with an initial value of 0.3 and adjusted in real time based on the noise intensity (range 0.1-0.6).

[0043] The graph convolutional network model of the entity relationship evaluation network uses two GCN layers, each with 256 hidden units, a ReLU activation function, and a mean squared error (MSE) loss function. Parameters are updated using a backpropagation algorithm. The concept mapping deviation rate threshold is set based on the business needs of different fields, typically between 0.1 and 0.3, and can be manually adjusted through the system management interface.

[0044] The feedback delay between the knowledge distillation optimization module and the policy optimization node does not exceed 50ms, ensuring the system's real-time response to deviations. The parameter update cycle of the cross-domain knowledge fusion model matches the detection frequency of the entity relationship evaluation network, once per second, to ensure that the system can promptly adapt to dynamic changes in data characteristics.

[0045] Example 2: This example describes the construction conditions for the fusion subnet and the multimodal switching link of the conceptual topology modeling unit. By setting strict fusion subnet construction conditions, the system ensures the effectiveness of the dynamic dimensional fusion network in the heterogeneous data acquisition module. Furthermore, the conceptual topology modeling unit uses the multimodal switching link to adaptively switch data processing paths in both steady-state and abnormal scenarios, ensuring the accuracy and robustness of semantic association analysis.

[0046] 1. Construction Conditions and Implementation Logic of the Converged Subnet The construction of the fusion subnet configured with multi-source data perception nodes must meet three conditions: the meta-attribute characteristics are within the preset domain range, the entity recognition rate corresponding to the fusion subnet is greater than the threshold, and the semantic association density is less than the allowable deviation range. The specific implementation is as follows: When structured data's meta-attribute features (such as the value range of numeric fields and the set of legal values ​​for enumerated fields) are connected to the system, they are first validated using pre-set domain interval validation rules. For example, the pre-set interval for the "transaction amount" field in financial data is [0, 106]. If a meta-attribute feature value captured in real time exceeds this interval (e.g., -500 or 1.2×107), it is considered invalid and the corresponding fusion subnet is not constructed. Latent semantic labels for unstructured data (e.g., entity type labels such as "person name" and "organization name" in text) must match the pre-set domain label set. If a label belongs to a domain-independent category (e.g., the label "weather" appears in medical data), it is considered out of the domain interval and does not trigger the creation of a fusion subnet.

[0047] The entity recognition rate corresponding to the fusion subnet is achieved through technologies such as natural language processing (NLP) and computer vision (CV). Taking text data as an example, the entity recognition module uses the BERT-BiLSTM-CRF model to perform named entity recognition (NER) on the input text and output entity labels such as names of people, places, and organizations. The formula for calculating the entity recognition rate is:

[0048] When the recognition rate is greater than the preset threshold (such as 85%), the fusion subnet is considered to have effective processing capabilities and is allowed to be built; if it is lower than the threshold, it means that the model's entity parsing ability for the current data is insufficient and needs to be optimized through data enhancement, model fine-tuning, etc. until the conditions are met.

[0049] Semantic association density is measured by summing the edge weights between concept nodes in the conceptual topology modeling unit, reflecting the closeness of semantic associations in the data. The permissible deviation range is pre-set based on domain characteristics. For example, the semantic association density of medical data is typically higher than that of social media data. When the real-time calculated semantic association density exceeds the permissible deviation upper limit (e.g., 1.5 times the normal range mean), it may indicate abnormal data aggregation (e.g., noise interference or data errors), and in this case, the fusion subnet will not be constructed. If it is below the permissible deviation lower limit (e.g., 0.5 times the normal range mean), the data semantic associations are considered too sparse, and constructing a fusion subnet is of limited significance, so it will not be created.

[0050] The construction process of the fusion subnet is as follows: after the multi-source data perception node captures data features in real time, it first performs meta-attribute feature domain interval verification. If it passes, it triggers the entity recognition rate calculation. If the recognition rate meets the standard, it then performs semantic association density deviation verification. When all three conditions are met, the dynamic dimension fusion network creates a new fusion subnet and connects it in series to the existing network to participate in data processing in the order of domain thresholds.

[0051] 2. Multimodal Switching Links of Conceptual Topological Modeling Units The multimodal switching link includes a steady-state association sub-link and an abnormal blocking sub-link, both of which realize adaptive switching of data processing paths through different trigger mechanisms and processing logic.

[0052] The steady-state association sub-link is generated based on the matching degree between the entity relationship signal of the semantic association parsing module and the core decision benchmark framework. The specific process is as follows: the concept topology modeling unit converts data features into concept nodes and associated edges to form entity relationship signals (such as the association strength and association type between nodes); the core decision benchmark framework generated by the strategy optimization node includes preset association scopes, weight distribution rules, etc. The matching degree is calculated using the cosine similarity algorithm, and the formula is:

[0053] When the matching degree is greater than or equal to a preset threshold (e.g., 0.8), the scenario is determined to be steady-state, and the steady-state association sub-link is activated. This sub-link uses forward propagation logic, with data sequentially passing through the various fusion sub-networks of the dynamic dimensional fusion network, the concept topology modeling unit, and the knowledge vector clustering unit, ultimately outputting to the decision graph generation module, forming a complete semantic association resolution path. During this process, the steady-state association sub-link monitors the output data of each module in real time. If an anomaly is detected (e.g., a sudden increase in the variance of the output features of a fusion sub-network), a warning signal is sent to the abnormal blocking sub-link, but the path is not immediately switched to avoid misjudgments.

[0054] The abnormal blocking sub-link is constructed based on the correlation between the instantaneous conflict risk of the information extraction path and the data noise event. The instantaneous conflict risk is identified by calculating the weight difference of each node in the information extraction path. The formula is:

[0055] When the difference exceeds a threshold (e.g., 0.6), a transient conflict risk is determined. Data noise events are detected using a sliding window algorithm. For example, if abnormal data is received three or more times within five seconds, a noise event is triggered.

[0056] The triggering conditions for abnormal blocking of sub-links are: the matching degree is lower than the threshold (such as 0.5) and the instantaneous conflict risk difference is greater than 0.6; a data noise event is detected and lasts for more than 1 second; the steady-state associated sub-link warning signal accumulates for more than 3 times.

[0057] When a sub-link is blocked due to an exception, the system performs the following operations: Path switching: cuts off the data transmission of the stable associated sub-link and imports the data into the abnormal blocked sub-link; Conflict resolution: Using the preset rules in the exception blocking parameter library, the node weights of the conflicting paths are redistributed, for example, reducing the weights of high-difference nodes and increasing the proportion of low-difference nodes. Noise filtering: Use the dynamic dimensional projection model to perform secondary feature extraction on the data and filter out the characteristic components corresponding to the noise. For example, principal component analysis (PCA) can be used to remove the principal components containing noise. Path verification: After switching, the output data of the abnormal blocking sub-link is monitored in real time. If the matching degree rises to above 0.7 for 5 consecutive cycles (the cycle length is configurable, the default is 100ms), the steady-state associated sub-link is gradually switched back. If the standard is not met, the abnormal blocking mode is maintained and the parameter update process of the cross-domain knowledge fusion model is triggered (such as the abnormal blocking instruction described in Example 4).

[0058] 3. Collaboration Mechanism and Technical Details of Multimodal Handover Links The stable associated sub-link and the abnormal blocking sub-link are mutually exclusive managed through the switching control module. This module adopts the finite state machine (FSM) design. The states include: stable operation, abnormal warning, blocking switching, and recovery verification. The state transition logic is as follows: Steady-state operation → Abnormal warning: When the matching degree drops below 0.8 but above 0.5, or the instantaneous conflict risk difference is between 0.4 and 0.6, the system enters the abnormal warning state and initiates lightweight data verification (e.g., sampling and checking 10% of the data features). Abnormal warning → blocking switching: If two consecutive verification failures occur in the warning state (for example, the sampling data error rate exceeds 15%), or the matching degree is lower than 0.5, the blocking switching is triggered and the abnormal blocking sub-link is entered; Blocking switching → recovery verification: After the abnormal blocking sub-link runs for 5 cycles, it automatically enters the recovery verification state to detect the matching degree and conflict risk indicators; Recovery verification → Steady-state operation / blocking switch: If the indicators meet the standards, return to steady-state operation; otherwise, maintain the blocking switch state and record the abnormal log.

[0059] In terms of technical implementation, the multimodal switching link uses a lock-free queue (Lock-FreeQueue) to achieve zero-copy switching of the data path, ensuring switching latency of less than 50 microseconds. The stable association sub-link and the abnormal blocking sub-link are each configured with independent computing resources (such as GPU cores and memory space) to avoid processing delays caused by resource competition. The conceptual topology modeling unit is shared by both links, but differentiated processing is achieved through different parameter configurations (for example, high-precision parsing parameters are used for the stable link and anti-interference parameters are used for the abnormal link).

[0060] IV. Application Scenarios of Converged Subnet Construction and Multimodal Switching Take the power system feasibility study data processing as an example: Fusion subnet construction: When capturing real-time current data from smart meters, the meta-attribute features (e.g., current value range 0-100A), entity recognition rate (device number recognition rate must be ≥90%), and semantic association density (the association density between current, voltage, and power must be within the normal operating range) are first verified. Once these conditions are met, a "current feature fusion subnet" is created and connected in series to the dynamic dimension fusion network. This subnet is responsible for feature extraction and fusion of this data.

[0061] Multimodal switching: When a power grid fault causes a significant amount of impulse noise in the data, the matching degree of the steady-state association sub-link rapidly drops to 0.4, while the instantaneous conflict risk difference rises to 0.7, triggering the abnormal blocking sub-link. After the system switches to abnormal mode, it applies the fault data processing rules in the abnormal blocking parameter library to enhance impulse noise filtering, redistribute information extraction path weights, and prioritize the analysis of key fault-related features (such as voltage sag nodes), ensuring the accuracy of fault diagnosis decisions.

[0062] Example 3: This example details the steps for generating a cross-domain knowledge fusion model and a core decision benchmark framework for the policy optimization node. The cross-domain knowledge fusion model configured in the policy optimization node integrates attribute-semantic mappings and anomaly blocking parameters for heterogeneous data, enabling unified representation and analysis of multi-source data. The core decision benchmark framework is generated based on the dynamic fusion of real-time data and historical knowledge, ensuring the accuracy and robustness of decision logic through multi-stage calculation and correction.

[0063] The cross-domain knowledge fusion model, a core component of the strategy optimization node, primarily consists of an attribute-semantic mapping table, an anomaly blocking parameter library, and a fusion inference engine. The attribute-semantic mapping table utilizes a multi-level index structure to associate and map meta-attributes of heterogeneous data with semantic labels. For example, in the financial sector, the "transaction amount" attribute in structured data can be mapped to the semantic labels "large transaction" and "small transaction." The mapping rules are based on preset thresholds (e.g., an amount > 1 million is considered a large transaction). Natural language processing techniques are used to extract keywords from text descriptions in unstructured data (e.g., "buy stocks" and "redeem funds") and map them to the semantic category "investment behavior." The mapping table supports dynamic updates, automatically expanding the mapping relationship through an incremental learning algorithm when new attribute features or semantic labels are detected.

[0064] The exception blocking parameter library stores the system's accumulated exception handling experience over the course of historical operations, including characteristic patterns, blocking strategies, and parameter configurations for various exception scenarios. For example, when an unusual fluctuation in data is detected (such as a surge in transaction frequency within a short period of time), the corresponding exception handling strategy in the parameter library is activated. This strategy may include actions such as reducing the weight of relevant information extraction paths and increasing the frequency of data verification. The parameter library utilizes a hierarchical storage structure, categorizing and indexing exceptions by type (e.g., data noise, logical conflicts) and severity (e.g., minor, moderate, severe) for rapid retrieval and application.

[0065] The fusion inference engine is implemented based on a graph neural network architecture. It uses attribute-semantic mapping tables and anomaly blocking parameter libraries as prior knowledge to perform joint inference on real-time input data. Specifically, the engine converts heterogeneous data into a knowledge graph representation, where nodes represent entities (such as users and transaction records) and edges represent relationships between entities (such as "user-initiate-transaction"). A graph convolutional network (GCN) is used to extract and propagate features from the knowledge graph, learning the latent semantic connections between entities. Furthermore, the fusion inference engine incorporates an attention mechanism that dynamically adjusts the focus on different entities and relationships based on information in the anomaly blocking parameter library, enhancing robustness against anomalous data.

[0066] The generation of the core decision benchmark framework is based on the output of the cross-domain knowledge fusion model and is achieved through multi-stage calculation and correction. The specific steps are as follows: Step 1: Calculate the theoretical upper bound of the association range for the steady-state association sublink. Based on the real-time attribute data and weight ranking results from the heterogeneous data acquisition module, calculate the theoretical upper bound of the association range for each level of the fusion subnet in the steady-state association sublink. This upper bound reflects the degree of data association that the fusion subnet can effectively handle under normal business logic. The calculation formula is:

[0067] in, Indicates the The theoretical upper limit of the correlation range of the level fusion subnet, is the maximum correlation value in the historical processing data of the subnet, is the historical average correlation value, is an adjustment factor (ranging from 0.6 to 0.8) used to balance the impact of extreme situations with the impact of normal conditions. This formula dynamically determines the reasonable boundaries of the correlation range by combining historical statistical data with real-time adjustments.

[0068] Step 2: Generate a decision logic reference graph. The upper limit of the theoretical correlation range calculated in Step 1 and transient data noise are input into the cross-domain knowledge fusion model. Based on the attribute-semantic mapping table and the anomaly blocking parameter library, the fusion inference engine jointly analyzes the input data to generate a decision logic reference graph. This graph is represented as a graph structure, with nodes representing decision elements (such as business indicators and risk factors) and edges representing logical relationships between elements (such as causal relationships and dependencies). Edge weights in the graph are dynamically adjusted using an attention mechanism to reflect the importance of different elements in the current decision scenario. For example, in a financial risk assessment scenario, when abnormal market fluctuations are detected, the weights of edges related to market risk will increase significantly, highlighting their impact on the decision.

[0069] Step 3: Conflict Risk Correction in the Abnormal Blocking Sublink: The abnormal blocking sublink is used to correct conflict risks in the decision logic reference graph. Specifically, the system first detects whether there are conflicting paths in the reference graph (e.g., two paths leading to contradictory decision conclusions). If so, corrections are made based on the conflict resolution strategies in the abnormal blocking parameter library. For example, if a user's transaction behavior is detected to trigger both the "high-risk" and "normal" decision paths, the system will invoke the priority rules for transaction risk assessment in the abnormal blocking parameter library, prioritize the more reliable risk indicator path, and adjust the reference graph accordingly. Furthermore, the abnormal blocking sublink considers the impact of data noise events and weights decision factors that are most affected by noise, ensuring the high reliability of the resulting core decision benchmark framework.

[0070] The cross-domain knowledge fusion model and the core decision benchmark framework generation process collaborate through a data exchange and feedback mechanism. Regarding data exchange, the output of the cross-domain knowledge fusion model (such as the decision logic reference graph) directly serves as input for the core decision benchmark framework generation, ensuring that the decision logic is based on a comprehensive knowledge representation. Furthermore, any new anomaly patterns or parameter adjustment requirements discovered during the core decision benchmark framework generation process are fed back to the cross-domain knowledge fusion model, prompting it to update the attribute-semantic mapping table and anomaly blocking parameter library, forming a closed-loop optimization loop.

[0071] In terms of feedback mechanisms, the system has established a multi-level feedback channel. When the core decision benchmark framework deviates in actual application, the first level of feedback is triggered, and local optimization is performed by adjusting the policy parameters in the abnormal blocking parameter library. If the deviation persists, the second level of feedback is triggered, initiating the update process of the attribute-semantic mapping table, and expanding or correcting the mapping relationship through incremental learning algorithms. In addition, the system also has a periodic global feedback mechanism to regularly conduct comprehensive evaluation and updates of the cross-domain knowledge fusion model and the core decision benchmark framework, ensuring that the system always adapts to the ever-changing data environment.

[0072] The graph neural network of the cross-domain knowledge fusion model uses a multi-head attention mechanism with 8 heads to capture semantic associations across different dimensions. Model training combines supervised and self-supervised learning, leveraging historically annotated data for supervised training while simultaneously exploring the underlying semantic structure in the data through contrastive learning. Training uses the Adam optimizer with a learning rate of 0.001 and a batch size of 64. The number of training epochs is dynamically adjusted based on the data size.

[0073] Adjustment coefficients in the generation step of the core decision benchmark framework Adaptive adjustment strategy is adopted, and the initial value is set to 0.7. The system monitors the matching between the upper limit of the theoretical correlation range and the actual data correlation degree in real time. When the matching error exceeds 10% for 5 consecutive cycles, the gradient descent algorithm is used to fine-tune the The value is adjusted with a step size of 0.01 until the matching error drops to an acceptable range.

[0074] Conflict risk correction for abnormally blocked sub-links uses a heuristic search algorithm with a search depth of three layers to balance correction effectiveness and computational efficiency. During the correction process, the system maintains a conflict resolution log, recording the cause, strategy, and effect of each correction for subsequent analysis and optimization.

[0075] Example 4: In the production management system of a smart manufacturing enterprise, the anomaly blocking mechanism of a cross-domain knowledge fusion model plays a key role. The system collects heterogeneous information from multiple sources, including production equipment data, material inventory data, and process parameter data, in real time to construct a decision-making graph to optimize production processes. When the system detects data anomalies, it triggers different levels of anomaly blocking instructions according to pre-set rules, ensuring decision accuracy and system stability.

[0076] Triggering and responding to the first exception blocking instruction: At 10:00 AM one day, while analyzing equipment operating data, the system discovered that the "temperature" meta-attribute characteristic value for a key piece of production equipment reached 180°C, exceeding the preset range of [50°C, 150°C]. Simultaneously, the concept mapping deviation rate output by the concept topology modeling unit was 0.35, exceeding the 0.25 threshold. The simultaneous fulfillment of these two conditions immediately triggered the first exception blocking instruction.

[0077] The system quickly activated the alternative knowledge unit, which stores backup strategies for handling equipment anomalies. Simultaneously, the semantic association analysis module adjusted its granularity from minute-level analysis to second-level analysis to more accurately capture changes in equipment status. Specifically, the system temporarily deactivated the knowledge model based on normal operating conditions and instead invoked the equipment anomaly diagnosis model, which specifically analyzes equipment overtemperature conditions. The semantic association analysis module began analyzing equipment sensor data on a second-by-second basis, identifying temperature trends and fluctuations in related parameters.

[0078] Over the next five minutes, the system executed the first exception blocking instruction and continued to monitor the equipment temperature. It found that while the temperature exceeded the normal range, it remained stable and did not continue to rise. At this point, the system determined that the equipment was likely experiencing a temporary overload, not yet a serious failure. Therefore, the system maintained the current production plan but increased monitoring of the equipment.

[0079] The second exception blocks the triggering and execution of the instruction: If the device temperature does not return to the acceptable range within 10 minutes after the first abnormal blocking instruction is executed, but instead continues to rise to 200°C, the system will trigger the second abnormal blocking instruction. At this point, the multimodal switching link immediately switches from the stable association sub-link to the abnormal blocking sub-link, redistributing the topological weights of the information extraction path.

[0080] The system dynamically adjusts the priority of each information extraction path based on the duration of the data noise event (here, 10 minutes). The previously highly weighted "production progress" path has its weight reduced, while the weight of "equipment safety"-related paths has been significantly increased. For example, during the decision map generation process, information regarding equipment maintenance and safety warnings is prioritized, while information regarding production plan adjustments is temporarily shelved.

[0081] At the same time, the system adjusts the parameters of the abnormally blocked sub-links based on the characteristics of data noise events. In this case, due to abnormal device temperature, the system increased the sampling frequency of temperature sensor data from 1 to 5 times per second and adjusted the correlation analysis weights of other relevant parameters (such as pressure and vibration frequency) to more comprehensively assess the device status.

[0082] Triggering and responding to the third abnormal blocking instruction: When the entity recognition rate of a fusion subnet at a certain level in the heterogeneous data acquisition module falls below a threshold, a third exception blocking instruction is triggered. For example, when analyzing raw material quality inspection data, the entity recognition rate of the fusion subnet responsible for identifying "metal impurity content" drops to 75%, below the 85% threshold. The system immediately blocks the associated output channels of this subnet to prevent erroneous data from influencing decision-making.

[0083] At the same time, the periodic fluctuation suppression unit diverts the corresponding data stream to another fusion subnet. In this case, the "metal impurity content" test data is rerouted to the backup "material composition analysis" fusion subnet. This backup subnet uses different algorithms and feature extraction methods to perform a secondary analysis of the same batch of raw materials to ensure data accuracy.

[0084] During the execution of the third abnormal blocking instruction, the system continuously monitors the load of each fusion subnet. If other fusion subnets show signs of overload due to the additional data flow (for example, processing delay exceeds 200ms), the fourth abnormal blocking instruction is triggered.

[0085] Implementation and downgrade processing of the fourth abnormal blocking instruction: In the above example, when the backup "Material Composition Analysis" fusion subnet became overloaded due to processing additional data streams, the system triggered the fourth abnormal blocking instruction. The knowledge distillation optimization module immediately downgraded the output strategy of the rule inference engine and limited the association requirements of the edge domain.

[0086] The system first constructs a domain-related security level table, dividing each domain in the production management system into core domains (such as equipment safety and product quality) and peripheral domains (such as energy consumption statistics and staff scheduling). In situations of resource constraints, priority is given to ensuring the associated needs of core domains. In this case, "equipment safety" and "product quality" are classified as core domains, while "energy consumption statistics" is classified as a peripheral domain.

[0087] The rule-based inference engine dynamically downgrades the priority of associated paths based on the security level. The previously high-priority "energy optimization" path is downgraded, ensuring that resources are focused on analyzing equipment anomalies and raw material quality issues. Simultaneously, the semantic association parsing module's analysis frequency is reduced from 10 to 3 times per second to match the downgraded decision-making strategy and reduce system burden.

[0088] The system also activated a resource scheduling mechanism, temporarily allocating additional computing resources (such as GPU cores and memory space) to key fusion subnetworks to ensure they could efficiently process data in core areas. In this case, the "Material Composition Analysis" fusion subnetwork received additional computing resources, enabling it to maintain a high entity recognition rate despite the increased load.

[0089] Coordination and recovery mechanism of abnormal blocking instructions Throughout the entire exception handling process, each level of exception blocking instructions does not operate independently, but rather works in tandem to form a hierarchical response system. For example, after the fourth exception blocking instruction is triggered, the system will continue to monitor equipment temperature and raw material quality data. If the situation deteriorates, it may further adjust the handling strategy, such as suspending the relevant production process.

[0090] Once the abnormality is alleviated, the system gradually removes the blocking instructions according to the pre-set recovery mechanism. For example, if the device temperature is abnormal, after the temperature returns to the normal range and stabilizes for 15 minutes, the system first removes the second abnormal blocking instruction and switches back to the stable associated sub-link. Then, after a period of observation to confirm that the device is operating normally, the first abnormal blocking instruction is removed, restoring the normal knowledge model and parsing granularity.

[0091] Throughout the entire exception handling process, the system records detailed logs, including exception type, trigger time, handling measures, recovery time, and other information. This log data is used for subsequent system optimization and problem tracing. By analyzing historical exception cases, we continuously improve the exception blocking parameter library and handling strategies, thereby enhancing the robustness and adaptability of the system.

[0092] Through this multi-level exception blocking mechanism, the production management system of the intelligent manufacturing enterprise can effectively respond to various data anomalies, ensure the accuracy of decision-making and the stability of production, and maintain efficient operation even in complex and changing production environments.

[0093] Example 5: This example describes the supplementary steps for building a fusion subnet and generating a decision logic reference graph. The system optimizes the fusion subnet topology using a domain gradient adaptation index table and a semantic weighted regression algorithm. It also combines a knowledge graph database with anomaly interference layers to generate a precise decision logic reference graph. The following describes a detailed implementation using a product data mining scenario in the e-commerce sector.

[0094] During the construction of the M fusion subnetworks of the dynamic dimension fusion network, in addition to basic feature capture and weight allocation, the system generates a domain gradient adaptation index table based on the data clustering results of meta-attribute features and the fluctuation range of historical semantic labels. Taking e-commerce product data as an example, meta-attribute features include "product category", "price range", "sales level", etc. The product data is divided into high-end, mid-end, and low-end market clusters through the K-means clustering algorithm. The fluctuation range of historical semantic labels is determined by analyzing the frequency and time distribution of sentiment words (such as "high cost performance" and "poor quality") in user reviews. For example, the fluctuation of the "cooling effect" label of the "air conditioner" category in summer is significantly higher than in other seasons.

[0095] The domain gradient adaptation index table stores the clustering results and fluctuation characteristics corresponding to each fusion subnetwork in a tabular form, for example:

[0096] Based on the differences in relevance compensation coefficients between the fused subnets in the index table, the system uses a semantically weighted regression algorithm to assign topological weights to the feature transfer channels of the M fused subnets. For example, subnet A (high-end market) has the highest relevance compensation coefficient for the "brand" label, and its feature transfer channel is given a higher weight when processing the "brand-user loyalty" association path, ensuring that this type of high-value semantic information is transmitted first. At the same time, the system establishes redundant association links between fused subnets. For example, a "cross-level association channel" is set up between subnets A and B. When mid-range market characteristics appear in high-end market data (such as a high-end brand launching an affordable product line), the redundant link is automatically activated, avoiding semantic gaps caused by data isolation between subnets.

[0097] When generating a decision logic reference graph, the system first inputs the theoretical correlation range upper limit and transient data noise into the cross-domain knowledge fusion model, and simultaneously loads the preset knowledge graph database and data anomaly interference layer. Taking the e-commerce promotion decision-making scenario as an example, the knowledge graph database contains historical correlation data for "users, products, and promotions." For example, the combination of "young users + beauty products + holiday promotions" generally has a high conversion rate. The data anomaly interference layer records abnormal patterns that occurred in historical promotions. For example, "high discounts + low sales" may be caused by insufficient inventory or inadequate promotion.

[0098] Based on the ontological relationship strength data output by the conceptual topology modeling unit (e.g., the correlation strength between "product sales" and "promotion intensity" is 0.92), the system corrects the conflict threshold of the information extraction path in the reference graph. For example, when the "promotion intensity" and "profit margin" paths conflict (high discounts may lead to low profits), the system automatically increases the conflict threshold of the "profit margin" path based on the ontological relationship strength, giving priority to protecting the company's profit goals. In the abnormal blocking sub-link, the system superimposes the data sparsity parameter for secondary calibration. If the user feedback data for a promotion activity is sparse (e.g., the number of participants is less than 100), the system reduces the decision weight of the path and marks it as "needs to supplement data" to avoid decision-making bias caused by insufficient data.

[0099] Specific example: Fusion subnetwork and graph generation for summer air conditioner promotion decision-making E-commerce platforms need to develop promotional strategies for summer air conditioners. The data involved includes the price, sales volume, user reviews, inventory status, etc. of air conditioners of various brands. The data types include structured (price, sales volume) and unstructured (user reviews, product descriptions).

[0100] Using the K-means algorithm, the air conditioner data was divided into three clusters: high-end (unit price > 5,000 yuan), mid-range (3,000-5,000 yuan), and low-end (< 3,000 yuan), generating corresponding subnetworks A, B, and C. Historical semantic label analysis showed that the "quiet effect" and "intelligent control" labels for high-end air conditioners fluctuated by 40% during the summer months of June to August, while the "cost-effectiveness" label for mid-range air conditioners fluctuated by 25%, and the "promotional price" label for low-end air conditioners fluctuated by 35%. This information was used to populate the "fluctuation period" and "correlation compensation coefficient" fields in the domain gradient adaptation index table.

[0101] According to the index table, the correlation compensation coefficient of the "quiet effect" label of subnet A is 0.88, and the weight of its feature transmission channel is set to 0.9 when processing the "user evaluation-product selling point" path; a redundant link is established between subnet B and subnet C. When an inventory backlog of mid-range air conditioners is detected, some data can be diverted to the "promotional activities" processing path of subnet C to try to clear the inventory through a low-price strategy.

[0102] Based on historical data, the theoretical upper limit of the associated range for "promotion intensity" for Subnet A is set at 15% (high-end brands typically offer discounts of no more than 15%), 25% for Subnet B, and 35% for Subnet C. If an outlier value of 20% discount for Subnet A appears in real-time data, the system will flag it as transient data noise.

[0103] The knowledge graph database shows that the conversion rate of the "high temperature warning area + smart air conditioner" combination in historical promotions reached 30%, significantly higher than the average level; the data anomaly interference layer indicates that there were user complaints caused by "high-discount air conditioners out of stock" in the same period last year, and a conflict threshold needs to be set for the corresponding path.

[0104] Conflict Threshold Correction and Recalibration: The conceptual topology modeling unit outputs a "stock status - promotion feasibility" ontology relationship strength of 0.95. The system raises the conflict threshold for this path from the default 20% to 30%, allowing inventory deviations within 30% while still maintaining the promotion plan. Because the current inventory data is complete (sparseness <5%), no weight adjustment is performed on the abnormal blocking sub-link, and the reference graph directly enters the core decision benchmark framework generation phase.

[0105] The domain gradient adaptation index table is updated in sync with the data clustering results, with a default full update performed daily at midnight. Real-time incremental updates are triggered when a sudden change in the new data distribution (such as the addition of a new product category) is detected. The semantically weighted regression algorithm uses an iterative weighting approach. After processing each batch of data, the weight coefficient is automatically adjusted based on the path processing effect, with the adjustment range not exceeding 10% of the current weight to avoid excessive fluctuations.

[0106] The knowledge graph database, which serves as the reference graph for decision logic, supports multi-source data access, including internal business system data and third-party market data (such as weather data and competitor product prices). This data is cleansed and standardized through a unified data center. The data anomaly interference layer is constructed using a combination of crowdsourcing annotation and automatic detection. Operators can manually mark typical anomaly cases, and the system automatically identifies new anomaly patterns using machine learning algorithms (such as the Isolation Forest algorithm).

[0107] By default, redundant links between converged subnets are dormant and activate only when the processing latency of the primary link exceeds a threshold (e.g., 500ms) or the data feature matching falls below a preset value (e.g., 30%). Once activated, the redundant links operate asynchronously, avoiding the need to preempt computing resources from the primary link and ensuring overall system efficiency.

[0108] Through the domain gradient adaptation index table and semantic weighted regression algorithm, the system realizes the dynamic optimization of the fusion subnet topology structure and improves the processing accuracy of segmented domain data; the decision logic reference map integrates multi-source knowledge and abnormal experience to ensure that the generated decision framework conforms to historical laws and can adapt to real-time data changes, providing a technical solution with both accuracy and flexibility for feasibility data mining in e-commerce and other fields.

[0109] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0110] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A feasibility study data intelligent mining system, characterized by: include: Heterogeneous data acquisition module, semantic association analysis module and decision graph generation module; The heterogeneous data acquisition module includes multi-source data perception nodes and a dynamic dimension fusion network; The semantic association analysis module includes a concept topology modeling unit and a knowledge vector clustering unit; The decision graph generation module includes a strategy optimization node and a rule reasoning engine; the multi-source data perception node is used to capture the meta-attribute characteristics of structured data and the potential semantic labels of unstructured data in real time; according to the data update frequency and entity association density, a multi-layer data feature topology is constructed through the dynamic dimension fusion network; the dynamic dimension fusion network is composed of M domain threshold-based fusion subnets connected in series; based on the real-time semantic response characteristics of the M fusion subnets, a core decision benchmark framework is generated through the strategy optimization node, and combined with the association constraint signal output by the knowledge vector clustering unit, the information extraction path is dynamically weighted and sorted through the rule reasoning engine.

2. The feasibility study data intelligent mining system according to claim 1, characterized in that: The system also includes a temporal feature correction module and a knowledge distillation optimization module; The timing feature correction module includes a periodic fluctuation suppression unit and a noise interference compensation unit; According to the generated core decision benchmark framework and dynamic weight ranking results, the data semantic association range is corrected in time series through the periodic fluctuation suppression unit, and the vector parameters of the semantic association parsing module are dynamically smoothed through the noise interference compensation unit; at the same time, the concept mapping deviation rate is detected in real time through the entity relationship evaluation network constructed by the knowledge distillation optimization module, and the detection results are fed back to the strategy optimization node and rule inference engine to perform closed-loop iteration on the decision framework.

3. The feasibility study data intelligent mining system according to claim 2, characterized in that: The periodic fluctuation suppression unit is composed of feature transmission channels corresponding to the fusion subnets connected in series in the dynamic dimension fusion network; the steps of constructing the multi-layer data feature topology include: Capture meta-attribute features and semantic label fluctuation signals, input them into the dynamic dimension projection model configured in each level of fusion subnet, generate data correlation compensation coefficients and store them in the corresponding fusion subnet; Based on the correlation compensation coefficients stored in fusion subnets at all levels and combined with the instantaneous change rate of entity relationships, the information extraction weights are distributed to M fusion subnets through the feature importance evaluation algorithm to form a dynamic feature matching topology.

4. The feasibility study data intelligent mining system according to claim 3, characterized in that: The conditions for constructing the fusion subnet configured by the multi-source data perception node include: the meta-attribute feature is within the preset domain range, the entity recognition rate corresponding to the fusion subnet is greater than the threshold, and the semantic association density is less than the allowable deviation range; The conceptual topology modeling unit is also configured with a multimodal switching link; the multimodal switching link includes a steady-state association sub-link and an abnormal blocking sub-link; the steady-state association sub-link is generated based on the matching degree between the entity relationship signal of the semantic association parsing module and the core decision reference framework; the abnormal blocking sub-link is constructed based on the association relationship between the instantaneous conflict risk of the information extraction path and the data noise event.

5. The feasibility study data intelligent mining system according to claim 4, characterized in that: The strategy optimization node is configured with a cross-domain knowledge fusion model; the cross-domain knowledge fusion model includes an attribute-semantic mapping table corresponding to the heterogeneous data acquisition module and an abnormal blocking parameter library of the semantic association analysis module; The steps of generating the core decision benchmark framework include: Based on the real-time attribute data and weight ranking results of the heterogeneous data acquisition module, the theoretical upper limit of the association range of each level of fusion subnet in the steady-state association sub-link is calculated; The upper limit of the theoretical correlation range and transient data noise are input into the cross-domain knowledge fusion model to generate a decision logic reference map, and the conflict risk of the reference map is corrected through the abnormal blocking sub-link to form a core decision benchmark framework.

6. The feasibility study data intelligent mining system according to claim 5, characterized in that: The parameter updating step of the cross-domain knowledge fusion model includes: When it is detected that the meta-attribute feature exceeds the preset domain range or the concept mapping deviation rate exceeds the threshold, the first abnormal blocking instruction is triggered, that is, the replacement knowledge unit is started and the parsing granularity of the semantic association parsing module is adjusted; If the associated range has not been restored to the allowed range after the execution of the first exception blocking instruction, the second exception blocking instruction is triggered, that is, switching to the exception blocking sub-link through the multimodal switching link, and reallocating the topological weight of the information extraction path based on the duration of the data noise event.

7. The feasibility study data intelligent mining system according to claim 6, characterized in that: The parameter updating step of the cross-domain knowledge fusion model further includes: When the entity recognition rate of a certain level of fusion subnet in the heterogeneous data acquisition module is lower than the threshold, the third abnormal blocking instruction is triggered, that is, the associated output channel of the subnet is blocked, and the corresponding data flow is transferred to other fusion subnets through the periodic fluctuation suppression unit; During the execution of the third exception blocking instruction, if other fusion subnets are detected to be overloaded, the fourth exception blocking instruction is triggered, that is, the knowledge distillation optimization module is called to downgrade the output strategy of the rule inference engine and limit the association requirements of the edge field.

8. The feasibility study data intelligent mining system according to claim 7, characterized in that: The degradation processing logic of the knowledge distillation optimization module includes: Based on the deviation rate data and knowledge vector clustering indicators output by the entity relationship evaluation network, a domain-related security level table is constructed; When the fourth abnormal blocking instruction is triggered, the priority of the associated path is dynamically downgraded according to the security level table, and the parsing frequency of the semantic association parsing module is synchronously adjusted to match the downgraded decision-making strategy.

9. The feasibility study data intelligent mining system according to claim 3, characterized in that: The step of constructing the fusion subnet further includes: Generate a domain gradient adaptation index table based on the data clustering results of meta-attribute features and the fluctuation range of historical semantic labels; According to the difference in correlation compensation coefficients of each fusion subnetwork in the index table, the semantic weighted regression algorithm is used to distribute topological weights to the feature transmission channels of the M fusion subnetworks, and redundant association links are established between the fusion subnetworks.

10. The feasibility study data intelligent mining system according to claim 5, characterized in that: The generation process of the decision logic reference map includes: After inputting the theoretical correlation range upper limit and transient data noise into the cross-domain knowledge fusion model, the preset knowledge graph database and data anomaly interference layer are loaded simultaneously; Based on the ontology relationship strength data output by the concept topology modeling unit, the conflict threshold of the information extraction path in the reference graph is corrected, and the data sparsity parameter is superimposed in the abnormal blocking sub-link for secondary calibration.

Images