Data acquisition device for engineering informatization management

Abstract

The invention relates to the technical field of engineering informatization management, and particularly provides a data acquisition device for engineering informatization management, which comprises a data network construction system for acquiring sensor data, environment monitoring data and manual input data of an engineering site, and standardizing the data into a uniform format through data preprocessing; constructing an initial data network; the data random sampling system identifies a multi-level data dependency relationship by comparing the KL divergence with a dynamic threshold value; a multi-level data fusion network is formed; according to the constructed multi-level data fusion network, random sampling is carried out on the network, and initial data representation is obtained; the engineering data representation system constructs a positive example pair and a negative example pair in the multi-level data fusion network, node representations in the multi-level data fusion network are aggregated, and a final engineering data representation is obtained. According to the method, the optimized data representation which not only can retain original data characteristics, but also can reflect the deep incidence relation is generated.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering information management, and in particular to a data acquisition device for engineering information management. Background Art

[0002] In modern engineering information management, data acquisition devices, as core components, undertake the critical task of acquiring real-time data from various engineering environments. This data is not only used for project progress monitoring, quality assessment, and resource scheduling, but also plays a vital role in decision support systems. However, existing data acquisition devices for engineering information management have significant shortcomings in adaptability to complex environments, data processing capabilities, and the accuracy of logical deduction. Especially in multi-dimensional and multi-layered data collection, existing devices struggle to achieve efficient data integration and analysis, leading to information lag and decision-making bias. Current technical solutions are typically based on simple sensor networks or single-function data acquisition modules. Their operating principle can be summarized as follows: raw data is collected by sensors, initially filtered, and then transmitted to a central processing unit for storage and display. However, this linear processing model exhibits significant limitations when dealing with complex engineering scenarios. Existing technologies often fail to accurately calculate weight coefficients, resulting in distorted data fusion results and affecting the reliability of subsequent analysis.

[0003] Prior art, Chinese patent application number 202410169778.9, discloses a digital archive management system and method for water conservancy project projects. The system includes the following steps: developing and establishing a dedicated database based on a B / S architecture; establishing a project information database; creating a project organization; establishing a project management process; organizing and compiling project data generated by all participating units during the project construction cycle into digital archives; digitizing the compiled and established digital archives, along with traditional paper-based digital documents and external project documents, into PDFs and uploading them to a dedicated database for archiving; and managing the archived archives. While this system can automatically generate and archive archives based on the collaborative work of participating units across different project operations during the water conservancy project construction process, it only enables digital storage and archiving of archives and fails to address the dynamic integration of multi-source heterogeneous data. The data sources are limited (primarily manual entry and document scanning), and the system lacks the ability to intelligently analyze and mine real-time sensor data and environmental monitoring data.

[0004] Prior art 2, Chinese patent application number 202311472759.5, discloses an intelligent, engineering information management system, including a project management platform unit, which is equipped with a real-time monitoring system unit, a remote monitoring and control unit, and a data management and analysis unit. The project management platform unit integrates and manages all aspects of the bridge construction project. The real-time monitoring system unit uses sensors and monitoring equipment to monitor the bridge construction site in real time. The monitoring equipment monitors and records the deformation, temperature, and vibration parameters of the bridge structure, as well as the safety conditions of the construction site. Although the system provides comprehensive safety assurance, refined management, and efficient monitoring for bridge construction projects through real-time monitoring, scientific decision-making, remote monitoring and control, and construction risk reduction, thereby improving construction quality and reducing the risk of accidents, it relies on real-time monitoring data (such as deformation and temperature), which is only used for safety warnings and basic analysis. A deep correlation network between the data is not established, making it impossible to adaptively optimize data representation.

[0005] Prior art three, Chinese patent application number 202410598573.2, discloses a BIM-based engineering information management system. This system uses a set partitioning module to display the construction information of each sub-project on the BIM interface during construction, obtain multiple factors that affect the sub-project's construction progress, and analyze the multiple factors of each sub-project using a regression analysis algorithm to generate a progress coefficient for each sub-project. A dynamic construction period adjustment module periodically obtains the progress coefficients of all sub-projects in the affected set, performs a weighted calculation on the progress coefficients of all sub-projects to obtain an adjustment index. This adjustment index is used to dynamically adjust the predicted construction period, and the dynamic adjustment results are sent to the BIM model for display. While this system can dynamically predict the construction period based on the status of each sub-project, effectively avoiding over-management or under-management, it relies on a preset model (such as linear weighting) to process multiple factors, lacks the ability to dynamically model nonlinear data relationships, and does not address data heterogeneity.

[0006] Currently, existing technologies 1, 2, and 3 have problems such as isolated multi-source data, insufficient dependency mining, and rigid data analysis. Therefore, the present invention provides a data acquisition device for engineering information management. Summary of the Invention

[0007] In order to achieve the above object, the present invention adopts the following technical solutions: One aspect of the present invention provides a data acquisition device for engineering information management, comprising: The data random sampling system is configured to identify multi-level data dependencies by comparing the KL divergence with the size of the dynamic threshold; based on the multi-level data dependencies, new nodes are created and connected through edges to form a multi-level data fusion network; based on the constructed multi-level data fusion network, random sampling is performed on the network to obtain the initial data representation.

[0008] In an optional embodiment, the data random sampling system includes: The dependency location subsystem is configured to identify the difference areas exceeding the threshold based on the comparison results of the KL divergence between adjacent layer nodes and the dynamic threshold, and form a location map of multi-level data dependency relationships; The driver node reconstruction subsystem is configured to create new nodes at the corresponding layer gaps in the initial data network, guided by the dependencies identified in the positioning graph. The new nodes are connected to the bottom-level original nodes and the higher-level nodes through bidirectional edges, forming a cross-layer compensation path. The path generation subsystem is configured to superimpose the cross-layer connection relationships of all new nodes onto the initial data network to build a multi-level data fusion network; The fusion network guidance subsystem is configured to define the probability distribution of random sampling on the multi-level data fusion network according to the weight ratio of the two types of paths; by traversing the intersection nodes of the two types of paths, it extracts sub-network fragments covering multi-level features and generates initial data representation.

[0009] In an optional implementation, the driving node reconstruction subsystem includes: A dependency mapping extraction module is configured to extract marked cross-layer dependency regions based on a multi-level data dependency location graph; The bidirectional feature synthesis module is configured to complementarily fuse the extracted low-level feature weight distribution fragments with the high-level feature missing descriptions to generate feature weight distributions for new nodes. The fusion rules preserve the original data details for the low-level fragments and provide abstract constraints for the high-level missing descriptions. It also encodes the bidirectional feature representation required for cross-layer dependencies. The cross-layer connection rule generation module is configured to determine the connection strength of the new node with the original node at the bottom layer and the higher-layer node based on the feature weight distribution of the new node. When connecting to the bottom layer, the strength is determined by the proportion of the bottom-layer fragments in the distribution; when connecting to the higher layer, the strength is determined by the need to repair the missing descriptions in the distribution of the higher layer. This generates edges with bidirectional weight attributes, forming the connection rules of the cross-layer compensation path. The path dynamic embedding module is configured to embed bidirectional weighted edges into the initial data network according to the connection rules, so that the new nodes can establish physical connections with the bottom and high-level nodes at the same time, and finally form a cross-layer compensation path.

[0010] In an optional implementation, the cross-layer connection rule generation module includes: The feature distribution decoupling analysis submodule is configured to take the new node feature weight distribution generated by the bidirectional feature synthesis module as input and decompose it into two independent components: the low-level feature retention component and the high-level constraint injection component; The bidirectional strength parameterization submodule is configured to map the two independent components after decomposition into connection strength parameters of the bottom connection strength and the high-level connection strength respectively; The weight attribute synthesis independently combines the two types of connection strength parameters into a bidirectional weight attribute, whose structure contains two independent dimensions: the weight value from the bottom layer to the new node and the weight value from the new node to the upper layer; The rule topology submodule is configured to construct two types of connection edges for new nodes in the initial data network based on bidirectional weight attributes: bottom-level connection edges and high-level connection edges; the combination of the two types of edges constitutes the complete connection rules of the cross-layer compensation path.

[0011] In an optional implementation, the bidirectional intensity parameterization submodule includes: The component value extraction unit is configured to take the two independent components of the feature distribution decoupling analysis submodule output, the low-level feature retention component and the high-level constraint injection component, as input and extract their quantized values respectively; the extraction rules are as follows: a strength mapping rule activation unit configured to input the extracted values into a preset strength conversion program to generate connection strength parameters; The parameter normalization encapsulation unit is configured to limit the two types of strength parameters within a preset range and perform normalization alignment so that: the bottom-level connection strength parameter represents the priority of detail feature transmission, and the high-level connection strength parameter represents the urgency of constraint correction.

[0012] In an optional implementation, the intensity mapping rule activation unit includes: The conversion rule loading subunit is configured to take the low-level feature preservation component value output by the component value extraction unit and the high-level constraint injection component value as input and activate a preset intensity conversion program; a dynamic range calibration subunit, configured to perform range constraints and behavior correction on the initial intensity parameters; The parameter semantic alignment subunit is configured to semantically associate the calibrated bottom-level connection strength parameters with the high-level connection strength parameters; the bottom-level connection strength parameters and the high-level connection strength parameters together constitute the final connection strength parameters.

[0013] In an optional implementation, the engineering data representation system includes: The positive and negative example construction module is configured to construct two types of positive and negative example pairs based on the cross-layer compensation path of the multi-layer data fusion network and the hierarchical progressive path of the initial data network; The hierarchical difference measurement module is configured to define hierarchical difference measurement rules based on positive example pairs and negative example pairs; perform fusion network difference measurement and initial network difference measurement; The dual-network collaborative optimization module is configured to input the results of the two types of fused network difference metrics and the initial network difference metrics into the loss function and perform simultaneous optimization: The optimization result aggregation module is configured to perform hierarchical feature superposition of the optimized multi-level data fusion network node representation and the initial data network node representation; and finally generate an engineering data representation that is both stable and adaptable.

[0014] In an optional embodiment, the positive example pairs in the positive and negative example construction module are generated by: selecting, in a multi-layer data fusion network, new nodes and bottom / high-layer node pairs connected by the same cross-layer compensation path; and selecting, in an initial data network, node pairs whose KL divergence difference between adjacent layer nodes is lower than a dynamic threshold. Negative example pair generation: In a multi-level data fusion network, randomly select a new node with no connection path and any high-level node pair; in the initial data network, select node pairs whose cross-layer KL divergence difference exceeds a dynamic threshold.

[0015] In an optional embodiment, the method further comprises: The data network construction system is configured to obtain sensor data, environmental monitoring data and manual input data from the engineering site, build an initial data network, and calculate the feature weight distribution of each node.

[0016] In an optional embodiment, the method further comprises: The engineering data representation system is configured to construct positive example pairs and negative example pairs in a multi-level data fusion network, aggregate node representations in the multi-level data fusion network, and obtain a final engineering data representation.

[0017] The present invention involves acquiring heterogeneous data from multiple sources to construct a multi-level data fusion network, generating an initial data representation through random sampling, optimizing data quality through logical deduction, and finally generating a final data representation through node aggregation. By combining multi-level data fusion with an adaptive optimization algorithm, the present invention captures complex, multi-level dependencies, improves data expressivity and robustness, and reduces reliance on manually labeled data. The method is suitable for scenarios such as construction, equipment operation, and environmental monitoring, demonstrating broad applicability and scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is a block diagram of the data acquisition device for engineering information management provided in Example 1 of the present invention; Figure 2 A block diagram of a data network construction system provided in Example 2 of the present invention; Figure 3 This is a block diagram of the data random sampling system provided in Example 3 of the present invention; Figure 4 This is a block diagram of the engineering data representation system provided in Example 8 of the present invention; Figure 5 A block diagram of the electronic device provided by the present invention; Figure 6 A block diagram of a computer-readable storage medium provided by the present invention. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present invention will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0020] In the following, the terms "first," "second," etc., are used for descriptive convenience only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature identified with "first," "second," etc., may explicitly or implicitly include one or more of such features. In the description of the present invention, unless otherwise specified, "plurality" means two or more.

[0021] In the present invention, unless otherwise clearly specified and limited, the term "connection" should be understood in a broad sense. For example, "connection" can be a fixed mechanical connection, a detachable mechanical connection, or an integrated one; or, "connection" can be a direct connection or an indirect connection through an intermediate medium. In addition, unless otherwise clearly specified and limited, the term "coupling" should be understood in a broad sense. For example, "coupling" can be a direct electrical connection, such as physical contact and electrical conduction between two components, or it can be understood as the electrical connection between different components in a circuit structure through a physical line that can transmit electrical signals, such as printed circuit board (PCB) copper foil or wire, so as to transmit electrical signals; or, "coupling" can be an indirect electrical connection between two components through an intermediate medium; or, "coupling" can be an electrical connection between two components in an airless / non-contact manner, such as electrical connection between two components using capacitive coupling to transmit electrical signals.

[0022] In an embodiment of the present invention, directional terms such as "up", "down", "left" and "right" may be defined including but not limited to the orientation relative to the schematic placement of the components in the drawings. It should be understood that these directional terms may be relative concepts, which are used for relative description and clarification, and may change accordingly according to changes in the orientation of the components in the drawings.

[0023] Example 1: like Figure 1 As shown, an embodiment of the present invention provides a data acquisition device for engineering information management, comprising: The data network construction system is configured to acquire sensor data, environmental monitoring data, and manual input data from the project site, standardize them into a unified format through data preprocessing, construct an initial data network, calculate the feature weight distribution of each node, calculate the feature weight distribution of nodes in the next higher layer, calculate the KL divergence between nodes in adjacent layers, and compare the difference between the two feature weight distributions; and set a dynamic threshold. The data random sampling system is configured to identify multi-level data dependencies by comparing the KL divergence with the dynamic threshold; based on the multi-level data dependencies, new nodes are created and connected through edges to form a multi-level data fusion network; based on the constructed multi-level data fusion network, random sampling is performed on the network to obtain an initial data representation; The engineering data representation system is configured to construct positive example pairs and negative example pairs in a multi-level data fusion network, and construct positive example pairs and negative example pairs in an initial data network; then, the nodes in the multi-level data fusion network and the nodes in the initial data network are optimized respectively by using the positive example pairs, negative example pairs and loss functions in the corresponding hierarchical networks to obtain optimized data representations; and the node representations in the multi-level data fusion network are aggregated to obtain a final engineering data representation.

[0024] In the above-mentioned embodiment, the data acquisition device of this embodiment achieves multi-level intelligent fusion and optimized representation of engineering data through modular collaborative work. The specific technical significance is as follows: Data standardization and network construction: The data network construction system converts heterogeneous engineering data (sensor data, environmental monitoring data, and manual input) into a unified node network format. A hierarchical feature weight calculation combined with a KL divergence dynamic threshold mechanism is used to establish an initial network topology that quantifies data associations. Dynamic dependency mining: The data random sampling system automatically identifies hierarchical data associations based on KL divergence comparisons. By dynamically adding nodes and edges, it constructs a multi-level fusion network that reflects the true dependencies of engineering data. Its random sampling mechanism preserves the statistical characteristics of the original data. Contrastive learning optimization: The engineering data representation system uses positive and negative example contrast learning between the two networks (initial network and fused network) and adopts a loss function to drive bidirectional optimization. This improves the feature discrimination of the initial network nodes and enhances the cross-level representation capabilities of the fused network nodes. Ultimately, through aggregation operations, it outputs a complete representation of the engineering data with complete feature expression.

[0025] In summary, this embodiment achieves the adaptive fusion of multi-source heterogeneous data at the engineering site. Through dynamic network construction and comparative learning mechanisms, it generates optimized data representations that can both retain the characteristics of the original data and reflect deep correlation relationships, providing a high-quality data foundation for subsequent engineering information management.

[0026] Example 2:

[0027] like Figure 2 As shown, based on Example 1, the data network construction system provided by the embodiment of the present invention includes: The weight distribution generation module is configured to perform feature importance analysis on each underlying node (sensor, environmental monitoring, and manual input data unit) in the initial data network based on standardized unified format data, and generate a feature weight distribution for each node, reflecting the contribution strength of the data contained in the node in specific dimensions (such as time sensitivity, spatial correlation, data fluctuation amplitude and other implicit features); The weight transfer aggregation module is configured to take the feature weight distribution of the bottom-level nodes as input, and nonlinearly combine the distributions of all nodes in the same logical layer through the hierarchical superposition rule to generate the feature weight distribution of the nodes in the next higher layer; The quantization and dynamic adjustment module is configured to compare the feature weight distributions of two adjacent layers of nodes in a probability space. The KL divergence calculation quantifies the structural differences between the two layers. This difference is used to determine the effectiveness of inter-layer feature transfer. If the difference exceeds a dynamic threshold (determined by historical data training), network structural adjustments (such as adding cross-layer connections or splitting redundant nodes) are triggered. If the difference is within the threshold, the current hierarchical structure is retained. This mechanism ensures that the network can adapt to the dynamically changing characteristics of engineering data.

[0028] In the aforementioned embodiment, the weight distribution generation module of this embodiment converts the underlying nodes (such as sensors) of heterogeneous data sources into quantifiable feature weight distributions through feature importance analysis, achieving standardized feature representation of the original data in dimensions such as time and space. The weight transfer aggregation module uses a nonlinear combination algorithm to recursively propagate the underlying feature weights upward according to logical hierarchies, forming a hierarchical feature representation system. This hierarchical transfer preserves the implicit characteristics of the original data (such as time sensitivity) while achieving dimensionality reduction and abstraction of the feature space. The quantification and dynamic adjustment module uses KL divergence calculation to establish a structural difference measure between adjacent hierarchies. Its core value lies in: when the difference value exceeds a threshold, it triggers network structure adjustments (such as cross-layer connections / node splitting) to cope with dynamic changes in engineering data; when the difference value is within the threshold, it maintains the current network topology to ensure structural stability.

[0029] In summary, this embodiment builds a flexible network structure that adapts to changes in data features through the hierarchical transmission and dynamic adjustment of feature weights. While maintaining the network's hierarchical logic, the system achieves continuous tracking of data dynamic features and structural optimization.

[0030] Example 3: like Figure 3 As shown, based on Example 1, the data random sampling system provided by the embodiment of the present invention includes: The dependency localization subsystem is configured to identify the difference areas exceeding the threshold based on the comparison results of the KL divergence between nodes in adjacent layers and the dynamic threshold, representing the cross-layer dependencies that are not fully expressed in the initial data network, and form a localization map of multi-level data dependencies; The driver node reconstruction subsystem is configured to create new nodes at the corresponding layer gaps in the initial data network, guided by the dependencies identified in the localization graph. The feature weight distribution of the new nodes is synthesized from two types of data: the feature weight distribution fragments directly associated with the dependencies in the bottom-level nodes, and the missing features exposed by the KL divergence differences in the higher-level nodes. The new nodes are connected to the original bottom-level nodes and the higher-level nodes through bidirectional edges, forming a cross-layer compensation path. The path generation subsystem is configured to superimpose the cross-layer connection relationships of all new nodes onto the initial data network to construct a multi-layer data fusion network. The topology of the multi-layer data fusion network consists of two types of paths: the hierarchical progressive paths of the original data network and the cross-layer compensation paths constructed by the new nodes. The fusion network guidance subsystem is configured to define the probability distribution of random sampling on the multi-level data fusion network based on the weight ratio of the two types of paths (the original path weight is determined by the initial feature distribution, and the compensation path weight is dynamically adjusted by the KL divergence difference value); by traversing the intersection nodes of the two types of paths, sub-network fragments covering multiple levels of features are extracted to generate the initial data representation.

[0031] In the aforementioned embodiments, this embodiment locates cross-layer dependencies that are not fully expressed in the initial data network by comparing KL divergence with a dynamic threshold (the dependency location subsystem). This addresses the problem of weakened correlation between high-level and low-level features caused by layer-by-layer propagation in traditional hierarchical networks. Based on the located dependencies, new nodes with synthetic feature weights are inserted between the layers of the initial network (the driver node reconstruction subsystem). Cross-layer compensation paths are constructed through bidirectional connections, allowing the network to simultaneously preserve both the original hierarchical progressive paths and the newly added cross-layer paths (the path generation subsystem). The fusion network guidance subsystem defines a probability distribution through weight assignment (static weights for the original paths and dynamic weights for the compensation paths). This ensures that the random sampling process not only covers the hierarchical features of the initial network but also prioritizes cross-layer compensation paths with large differences in KL divergence. The resulting subnetwork fragments contain both low-level details and high-level semantic features. While maintaining the basic structure of the initial data network, cross-layer dependency compensation and dynamic path selection enhance the representational capabilities of the random sampling results for multi-level feature fusion.

[0032] Example 4: Based on Example 3, the driving node reconstruction subsystem provided in this embodiment of the present invention includes: The dependency map extraction module is configured to extract cross-layer dependency regions based on the multi-layer data dependency location map. Each region contains two types of information: a fragment of the feature weight distribution of the original nodes at the bottom layer (calculated by the data network construction system) and a description of the feature missing of the higher-level nodes (exposed by comparing the KL divergence difference value with the dynamic threshold). The bidirectional feature synthesis module is configured to complementarily fuse the extracted low-level feature weight distribution fragments with the high-level feature missing descriptions to generate feature weight distributions for new nodes. The fusion rules preserve the original data details (such as the temporal fluctuation characteristics of sensor data) for the low-level fragments, while providing abstract constraints for the high-level missing descriptions (such as the spatial correlation requirements of environmental monitoring data). It also encodes the bidirectional feature representation required for cross-level dependencies. The cross-layer connection rule generation module is configured to determine the connection strength of the new node with the original node at the bottom layer and the higher-layer node based on the feature weight distribution of the new node. When connecting to the bottom layer, the strength is determined by the proportion of the bottom-layer fragments in the distribution; when connecting to the higher layer, the strength is determined by the need to repair the missing descriptions in the distribution of the higher layer. This generates edges with bidirectional weight attributes, forming the connection rules of the cross-layer compensation path. The path dynamic embedding module is configured to embed bidirectional weighted edges into the initial data network according to the connection rules, so that the new nodes can establish physical connections with the bottom and high-level nodes at the same time, and finally form a cross-layer compensation path.

[0033] In the above-mentioned embodiment, the dependency mapping extraction module of this embodiment separates the underlying feature weight fragments and high-level feature missing descriptions from the positioning map, and clarifies the specific requirements of cross-layer dependencies (the detailed features that need to be retained at the bottom layer and the abstract constraints that need to be supplemented at the top layer). The bidirectional feature synthesis module generates the feature weight distribution of the new node through complementary fusion rules: retaining the details of the underlying original data (such as temporal fluctuation characteristics) while embedding the abstract constraints missing at the top layer (such as spatial association requirements), so that the new node has bidirectional expression capabilities. The cross-layer connection rule generation module dynamically allocates the connection strength of the new node with the bottom and top layers based on the feature distribution of the new node: the bottom connection strength depends on the proportion of feature fragments, and the high-level connection strength depends on the missing repair requirements, ensuring the bidirectional adaptability of the compensation path. The path dynamic embedding module embeds the bidirectional weighted edges into the initial network, so that the new node establishes a physical connection with the cross-layer node, and ultimately forms a compensation path with quantifiable weights.

[0034] In summary, this embodiment inserts new nodes with bidirectional expression capabilities into the hierarchical gaps of the initial data network through feature fusion and dynamic connection rules. The strength of the connection is driven by cross-layer dependency requirements, thereby achieving compensatory reconstruction of dependency relationships at the physical level.

[0035] Example 5: Based on Example 4, the cross-layer connection rule generation module provided in this embodiment of the present invention includes: The feature distribution decoupling analysis submodule is configured to take the new node feature weight distribution generated by the bidirectional feature synthesis module as input and decompose it into two independent components: Bottom-level feature retention component: corresponds to the proportion of feature weight distribution fragments of the bottom-level original nodes in the distribution (such as the retention ratio of sensor time series fluctuation features); High-level constraint injection component: corresponds to the intensity of the repair demand described by the high-level missing information in the distribution (such as the coverage of the spatial correlation constraints of environmental monitoring); The bidirectional strength parameterization submodule is configured to map the two independent components after decomposition into connection strength parameters: Bottom-level connection strength: directly determined by the proportion of the bottom-level feature retention component. The higher the proportion, the stronger the connection strength, ensuring the integrity of the detailed features transferred to the new node; High-level connection strength: The demand strength value of the component injected by the high-level constraint is dynamically adjusted. The more urgent the demand, the higher the connection strength, ensuring the efficiency of the abstract constraint in modifying the new node. The weight attribute synthesis independently combines two types of connection strength parameters into a bidirectional weight attribute, whose structure contains two independent dimensions: the weight value from the bottom layer to the new node, and the weight value from the new node to the upper layer. The attribute enables each edge to have the ability to adjust the direction, forming an asymmetric connection rule. The rule topology submodule is configured to construct two types of connection edges for new nodes in the initial data network based on the bidirectional weight attributes: bottom-level connection edges, which point from the original bottom-level node to the new node, and the weight value inherits the bottom-level connection strength parameter; high-level connection edges, which point from the new node to the target high-level node, and the weight value inherits the high-level connection strength parameter; the combination of the two types of edges constitutes the complete connection rules of the cross-layer compensation path.

[0036] In the aforementioned embodiment, the feature distribution decoupling analysis submodule of this embodiment decomposes the feature weight distribution of the new node into a low-level feature retention component (e.g., the proportion of temporal fluctuation features) and a high-level constraint injection component (e.g., the strength of spatial correlation requirements), identifying two independent influencing factors. The bidirectional strength parameterization submodule maps these two components into connection strength parameters: the low-level connection strength is directly determined by the proportion of the feature retention component, ensuring the integrity of the original detailed features; the high-level connection strength is dynamically adjusted by the strength of the constraint injection component, ensuring the efficiency of abstract constraint modification. The weight attribute synthesis combines these two strength parameters into a bidirectional weight attribute, forming an asymmetric connection rule: the weight value from the low-level to the new node reflects the requirement for detailed feature preservation; the weight value from the new node to the high-level reflects the requirement for constraint modification. The rule topology submodule constructs two types of directed edges in the initial network based on the bidirectional weight attributes: low-level connection edges (original node → new node), whose weights inherit the low-level strength parameters; and high-level connection edges (new node → target high-level node), whose weights inherit the high-level strength parameters.

[0037] In summary, this embodiment decouples feature distribution and parameterizes connection strength to generate asymmetric connection rules with directional adjustment capabilities, so that the construction of cross-layer compensation paths can simultaneously meet the two-way requirements of preserving underlying details and correcting high-level constraints, ultimately forming a cross-layer network topology with quantifiable structure and directable function.

[0038] Example 6: Based on Example 5, the bidirectional strength parameterization submodule provided in this embodiment of the present invention includes: The component value extraction unit is configured to take the two independent components of the feature distribution decoupling analysis submodule output, the low-level feature retention component and the high-level constraint injection component, as input and extract their quantized values respectively; the extraction rules are as follows: The value of the low-level feature retention component is directly calculated by the proportion of its distribution fragment in the overall feature weight; the value of the high-level constraint injection component is dynamically calibrated by the product of its coverage and the degree of difference exposure; a strength mapping rule activation unit configured to input the extracted values into a preset strength conversion program to generate connection strength parameters; Bottom-layer connection strength parameter: Linearly amplify the value of the bottom-layer feature retention component so that every 1% increase in the retention ratio corresponds to a fixed gradient increase in connection strength, ensuring that the integrity of detail transmission increases monotonically with the retention ratio; High-level connection strength parameter: The value of the high-level constraint injection component is thresholded and segmented. When the demand intensity is below the critical value, it is converted to a linear relationship. When it exceeds the critical value, the exponential growth mode is activated to achieve a higher intensity response to urgent demands. The parameter normalization encapsulation unit is configured to limit the two types of strength parameters to a preset range (such as the range of 0-1) and perform normalization alignment so that the bottom-level connection strength parameter represents the priority of detail feature transmission, and the high-level connection strength parameter represents the urgency of constraint correction.

[0039] In the aforementioned embodiments, this embodiment establishes a dual-channel quantization system for the feature space by concurrently processing low-level feature preservation components and high-level constraint injection components. The low-level channel utilizes static weighted proportion analysis to maintain the integrity of basic features, while the high-level channel achieves contextual awareness of constraints through the dynamic range-dissimilarity product. Together, these two dimensions form complementary dimensions of feature expression. The linear amplification rule in the low-level channel ensures proportionality in feature transfer, forming a stable baseline for detail transmission. The threshold triggering mechanism (linear-to-exponential conversion) in the high-level channel constructs a nonlinear response curve for constraint strength, enabling the system to prioritize urgent tasks in resource allocation. The parameter interval mapping established by the normalization encapsulation layer unifies the two strength parameters into comparable dimensions: the low-level parameters are converted into feature selection probability distributions, while the high-level parameters are encoded as time-sensitive coefficients for constraint modification. This process creates a parseable trade-off in parameter space between the gradual nature of feature preservation and the sudden nature of constraint execution.

[0040] In summary, this embodiment achieves context-adaptive allocation of network resources through a dynamic strength adjustment mechanism while maintaining the stability of basic feature transmission. When the system detects abnormal fluctuations in high-level constraints, it automatically triggers resource reallocation strategies while maintaining the progressive optimization of low-level feature transmission. This design effectively solves the problem of dynamically balancing feature preservation and constraint satisfaction in traditional networks.

[0041] Example 7: Based on Example 6, the intensity mapping rule activation unit provided in this embodiment of the present invention includes: The conversion rule loading sub-unit is configured to use the low-level feature retention component value (such as the sensor time series fluctuation feature retention ratio value) and the high-level constraint injection component value (such as the environmental monitoring space correlation constraint requirement strength value) output by the component value extraction unit as input to activate the preset strength conversion program; Bottom layer conversion program: maps the bottom layer values into linearly increasing connection strength parameters, and increases the corresponding strength value by a fixed amount every time the retention ratio increases by 1%, thus forming the initial strength parameters of the bottom layer connection strength parameters; High-level conversion program: compares the high-level value with the preset critical value and triggers the segmented conversion logic: if the value is lower than the critical value, the strength parameter is generated in a linear proportion; if the value exceeds the critical value, the exponential growth mode is started to generate the strength parameter; the initial strength parameter of the high-level connection strength parameter is output; a dynamic range calibration subunit, configured to perform range constraints and behavior correction on the initial intensity parameters; Bottom layer intensity calibration: Limit the linearly growing bottom layer parameters to a preset range (e.g., 0-0.8) to prevent excessive amplification of details and interference with high-level structures. High-level strength calibration: Apply a smooth decay factor to exponentially growing high-level parameters to avoid strength fluctuations caused by sudden demand changes, while ensuring that the parameter value does not exceed a preset upper limit (such as 1.0); a parameter semantic alignment subunit configured to semantically associate the calibrated bottom-level connection strength parameters with the high-level connection strength parameters; the bottom-level connection strength parameters and the high-level connection strength parameters together constitute a final connection strength parameter; Bottom-level connection strength parameter: indicates the priority of detailed feature transmission. A larger value indicates that the underlying features such as sensor timing fluctuations need to be transmitted more completely to the new node. High-level connection strength parameter: indicates the urgency of constraint correction. A higher value indicates that high-level constraints such as environmental monitoring spatial association need to act on new nodes more quickly.

[0042] In the aforementioned embodiment, the bottom-level channel utilizes a fixed-slope linear growth model to ensure a strictly monotonic relationship between feature retention ratio and transmission strength, establishing a stable basic feature transmission channel. The top-level channel utilizes a critical-value-triggered piecewise function (linear / exponential) to achieve a nonlinear response to constraint strength, forming a mechanism for prioritizing emergency matters. The bottom-level parameter range constraint (0-0.8) prevents oversaturated transmission of detailed features, maintaining the stability of the network infrastructure. The attenuation factor design of the top-level parameters effectively suppresses parameter oscillations caused by sudden signal mutations, controlling overshoot while ensuring response speed. The calibrated dual-channel parameters form orthogonal control dimensions: the bottom-level parameters quantify the integrity requirements of feature transmission, while the top-level parameters encode the timeliness requirements of constraint execution. The Cartesian product of the two forms a decision plane for connection strength, providing a parseable basis for resource allocation within the network.

[0043] In summary, this embodiment achieves a dynamic balance between feature preservation and constraint satisfaction: while ensuring the continuous transmission of basic features, it establishes a gradient response capability for sudden constraint demands and prevents system overload by pre-setting the boundaries of the parameter space. This mechanism enables the network to have the ability to adaptively switch between stable operation and emergency response.

[0044] Example 8: like Figure 4 As shown, based on Example 1, the engineering data representation system provided by the embodiment of the present invention includes: The positive and negative example construction module is configured to construct two types of positive and negative example pairs based on the cross-layer compensation path of the multi-layer data fusion network and the hierarchical progressive path of the initial data network; Positive example pair generation: In a multi-layer data fusion network, new nodes and bottom-layer / high-layer node pairs connected by the same cross-layer compensation path are selected; in the initial data network, node pairs whose KL divergence difference between adjacent layer nodes is below a dynamic threshold are selected; Negative pair generation: In a multi-layer data fusion network, randomly select a new node with no connection path and any high-level node pair; in the initial data network, select node pairs whose cross-layer KL divergence difference exceeds a dynamic threshold; The hierarchical difference measurement module is configured to define hierarchical difference measurement rules based on positive example pairs and negative example pairs; perform fusion network difference measurement and initial network difference measurement; The fusion network difference metric includes: positive example pair difference value, which calculates the similarity between the feature weight distribution of the new node and the nodes at both ends of the compensation path (the higher the similarity, the smaller the difference value); negative example pair difference value, which calculates the conflict strength between the feature weight distribution of the new node and irrelevant high-level nodes (the stronger the conflict, the larger the difference value); The initial network difference metrics include: positive pair difference value, which measures the matching degree of feature weight distribution of nodes in adjacent layers within the KL divergence threshold; negative pair difference value, which quantifies the deviation degree of feature weight distribution of nodes across layers due to KL divergence exceeding the threshold; The dual-network collaborative optimization module is configured to input the results of the two types of fused network difference metrics and the initial network difference metrics into the loss function and perform simultaneous optimization: Fusion network optimization: Reduce the difference value of positive pairs to enhance the feature compatibility of new nodes with cross-layer compensation paths; increase the difference value of negative pairs to suppress the feature coupling of new nodes with irrelevant high-level nodes; Initial network optimization: Reduce the difference between positive pairs to repair feature gaps between adjacent layer nodes due to KL divergence differences; increase the difference between negative pairs to strengthen feature independence between nodes across layers to match dynamic thresholds; The optimization result aggregation module is configured to perform hierarchical feature superposition of the optimized multi-level data fusion network node representation and the initial data network node representation; ultimately generating a stable and adaptable engineering data representation; Bottom-layer nodes: retain detailed features after initial network optimization (such as sensor timing fluctuations); High-level nodes: Fusion compensation path injection constraint features (e.g., spatial correlation of environmental monitoring); New node: inherits the bidirectional balance feature after cross-layer optimization.

[0045] In the above embodiments, this embodiment establishes multi-level association rules and exclusion rules between data nodes through the systematic construction of positive and negative example pairs; while the initial network maintains the hierarchical progressive relationship, it introduces cross-layer association constraints by using compensation paths; the KL divergence dynamic threshold is used to realize the adaptive evaluation of the matching degree of adjacent layer features, and the conflict intensity of cross-layer node relationships is quantified by feature weight distribution analysis; the dual-network collaborative optimization mechanism simultaneously guarantees: the hierarchical feature continuity of the initial network, the cross-layer feature compatibility of the fusion network, and the loss function design simultaneously constrains the attraction of positive example pairs and the repulsion of negative example pairs; the hierarchical feature superposition strategy is implemented, the bottom layer retains the optimized fine-grained features, the high-level layer integrates the cross-layer constraint features, and the new node obtains a bidirectional balance feature; the final representation has both: the temporal stability of the initial network (controlled by KL divergence) and the spatial adaptability of the fusion network (achieved by compensation paths).

[0046] In summary, this embodiment achieves a dynamic balance between hierarchical constraints and cross-layer associations in engineering data representation. Through the dual constraint mechanism of positive and negative example pairs, the generated representation satisfies both feature continuity and environmental adaptability requirements.

[0047] Example 9: Based on Example 8, the dual-network collaborative optimization module provided in this embodiment of the present invention includes: The difference value normalization processing submodule is configured to take the four types of difference values output by the hierarchical difference measurement module as input: the difference value of the positive example of the fusion network (the similarity between the new node and the nodes at both ends of the compensation path); The network's negative example difference value (the intensity of the conflict between the new node and the irrelevant high-level node) is integrated; the initial network's positive example difference value (the matching degree within the KL divergence threshold of adjacent layer nodes); and the initial network's negative example difference value (the degree of deviation of the cross-layer KL divergence exceeding the standard). These difference values are normalized and mapped to a unified numerical range (such as 0-1) to eliminate the interference of dimensional differences on subsequent calculations. The weight dynamic allocation submodule is configured to assign dynamic weight coefficients to the four types of normalized difference values based on the dynamic thresholds set by the data network construction system and the cross-layer compensation path weight ratios of the multi-layer data fusion network: Fusion network positive example weight: positively correlated with the compensation path weight ratio. The higher the path weight, the higher the positive example optimization priority. Negative example weight of the fusion network: positively correlated with the degree of exposure of the KL divergence difference. The more significant the difference, the stronger the need for negative example suppression. Initial network positive example weight: negatively correlated with the dynamic threshold. The lower the threshold (the stricter the requirement), the higher the adjacent layer matching optimization weight. Initial network negative example weight: positively correlated with the magnitude of the cross-layer KL divergence excess. The greater the deviation, the greater the independence reinforcement weight. The dual-network loss coupling submodule is configured to group the weighted four-category difference values by network type and construct a dual-network coupling loss function: Fusion network loss term = fusion network positive example weight × positive example difference value - fusion network negative example weight × negative example difference value; (Goal: minimize positive example difference, maximize negative example difference); Initial network loss term = initial network positive example weight × positive example difference value - initial network negative example weight × negative example difference value; (Goal: minimize positive example difference, maximize negative example difference); Total loss function = fusion network loss term + initial network loss term; The stability correction force balancing submodule is configured to introduce a stability factor (the degree of detail feature retention after initial network optimization) and a correction force factor (the degree of constraint feature coverage of the fusion network compensation path) to balance the total loss function. This forms a dynamically adaptive engineering loss function, the output of which drives parameter adjustments in the dual-network collaborative optimization module. When the stability factor falls below a preset benchmark, the weight of the initial network loss term is increased. When the correction factor does not meet expectations, increase the contribution ratio of the fusion network loss term.

[0048] In the above embodiment, this embodiment eliminates the dimensional differences of the four types of difference values through normalization processing, establishing a unified numerical benchmark for subsequent weight allocation and loss calculation; at the level of dynamic weight configuration optimization, the weight allocation mechanism is implemented: the weight of the positive example of the fusion network is linked to the importance of the compensation path, the weight of the negative example of the fusion network is linked to the degree of KL divergence anomaly, the weight of the positive example of the initial network is inversely linked to the strictness of the dynamic threshold, and the weight of the negative example of the initial network is linked to the cross-layer deviation amplitude; at the level of loss function construction, the dual-network coupling loss function is implemented: the loss term of the fusion network promotes positive example similarity / suppresses negative example conflict, the loss term of the initial network enhances intra-layer matching / weakens cross-layer interference, and the total loss function maintains the linear additivity of the two losses; at the level of dynamic balance adjustment, the stability correction force balance mechanism is implemented: the initial network feature retention is monitored through the stability factor, the fusion network constraint coverage is evaluated through the correction force factor, and the two types of loss weights are dynamically adjusted according to the compliance of the factors; In summary, this embodiment achieves the standardized comparability of multi-source difference indicators, dynamic priority configuration of optimization targets, mathematical coupling expression of dual-network optimization requirements, and closed-loop feedback regulation of stability and correction force; ultimately forming an engineering optimization framework with adaptive balancing capabilities, ensuring that the network optimization process simultaneously meets the dual requirements of feature preservation and constraint injection. Figure 5 A block diagram is shown of an exemplary electronic device suitable for implementing embodiments of the present invention.

[0049] The electronic device may include a central processing unit / microprocessor / main control chip, etc.; a storage medium, coupled to the central processing unit / microprocessor / main control chip, etc., and storing computer-executable instructions therein, for performing the steps of each method of an embodiment of the present invention when executed by the processor.

[0050] The central processing unit / microprocessor / main control chip etc. may include but is not limited to, for example, one or more processors or microprocessors etc.

[0051] The storage medium may include, but is not limited to, for example, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, computer storage media (such as hard disk, floppy disk, solid-state drive, removable disk, CD-ROM, DVD-ROM, Blu-ray disc, etc.).

[0052] In addition, the electronic device may also include (but not limited to) a data bus, an input / output bus / external bus / device bus, a display, and input / output devices (eg, keyboard, mouse, speaker, etc.).

[0053] The central processing unit / microprocessor / main control chip etc. can communicate with external devices via an I / O bus via a wired or wireless network (not shown).

[0054] The storage medium may also store at least one computer-executable instruction for executing the various functions and / or method steps in the embodiments described in this technology when executed by a central processing unit / microprocessor / main control chip, etc.

[0055] In one embodiment, the at least one computer executable instruction may also be compiled into or constitute a software product, wherein one or more computer executable instructions are executed by a processor to perform the various functions and / or method steps in the embodiments described in the present technology.

[0056] Figure 6 A schematic diagram of a computer-readable storage medium according to an embodiment of the present invention is shown.

[0057] like Figure 6 As shown, a non-transitory computer-readable storage medium stores instructions, such as computer-readable instructions. When the computer-readable instructions are executed by a processor, the various methods described above can be executed. Non-transitory computer-readable storage media include, but are not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. Non-transitory non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. For example, the non-transitory computer-readable storage medium can be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions stored on the computer-readable storage medium, the various methods described above can be performed.

[0058] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interface, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0059] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0060] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0061] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for executing all or part of the steps of the various embodiments of the method of the present invention via a computer device (which can be a personal computer, server, or network device, etc.). The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0062] In practical applications, this method can be widely applied to engineering scenarios such as construction data, equipment operation data, and environmental monitoring data. For example, in construction scenarios, by collecting sensor data (such as temperature, humidity, and vibration) and environmental monitoring data (such as wind speed and rainfall) from the construction site, combined with manually input data (such as construction progress records), a multi-level data fusion network can be constructed. This method can capture the complex multi-level dependencies of the construction site, generate high-quality data representations, and provide reliable support for subsequent project management and data analysis.

[0063] In summary, this invention, through the combination of a multi-level data fusion network and a logical inference optimization algorithm, effectively addresses the inability of existing data acquisition devices to handle complex logical inferences and multi-dimensional correlated data, achieving efficient collection and high-precision representation of engineering data. Its operating principles and implementation process fully demonstrate the technical advantages and application value of this invention.

[0064] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A data acquisition device for engineering information management, characterized in that: Include: The data random sampling system is configured to identify multi-level data dependencies by comparing the KL divergence with the size of the dynamic threshold; based on the multi-level data dependencies, new nodes are created and connected through edges to form a multi-level data fusion network; based on the constructed multi-level data fusion network, random sampling is performed on the network to obtain the initial data representation.

2. The data acquisition device for engineering information management according to claim 1, characterized in that: Data random sampling system, including: The dependency location subsystem is configured to identify the difference areas exceeding the threshold based on the comparison results of the KL divergence between adjacent layer nodes and the dynamic threshold, and form a location map of multi-level data dependency relationships; The driver node reconstruction subsystem is configured to create new nodes at the corresponding layer gaps in the initial data network, guided by the dependencies identified in the positioning graph. The new nodes are connected to the bottom-level original nodes and the higher-level nodes through bidirectional edges, forming a cross-layer compensation path. The path generation subsystem is configured to superimpose the cross-layer connection relationships of all new nodes onto the initial data network to build a multi-level data fusion network; The fusion network guidance subsystem is configured to define the probability distribution of random sampling on the multi-level data fusion network according to the weight ratio of the two types of paths; by traversing the intersection nodes of the two types of paths, it extracts sub-network fragments covering multi-level features and generates initial data representation.

3. The data acquisition device for engineering information management according to claim 2, characterized in that: Drive node reconstruction subsystem, including: A dependency mapping extraction module is configured to extract marked cross-layer dependency regions based on a multi-level data dependency location graph; The bidirectional feature synthesis module is configured to complementarily fuse the extracted low-level feature weight distribution fragments with the high-level feature missing descriptions to generate feature weight distributions for new nodes. The fusion rules preserve the original data details for the low-level fragments and provide abstract constraints for the high-level missing descriptions. It also encodes the bidirectional feature representation required for cross-layer dependencies. The cross-layer connection rule generation module is configured to determine the connection strength of the new node with the original node at the bottom layer and the higher-layer node based on the feature weight distribution of the new node. When connecting to the bottom layer, the strength is determined by the proportion of the bottom-layer fragments in the distribution; when connecting to the higher layer, the strength is determined by the need to repair the missing descriptions in the distribution of the higher layer. This generates edges with bidirectional weight attributes, forming the connection rules of the cross-layer compensation path. The path dynamic embedding module is configured to embed bidirectional weighted edges into the initial data network according to the connection rules, so that the new nodes can establish physical connections with the bottom and high-level nodes at the same time, and finally form a cross-layer compensation path.

4. The data acquisition device for engineering information management according to claim 1, characterized in that: Cross-layer connection rule generation module, including: The feature distribution decoupling analysis submodule is configured to take the new node feature weight distribution generated by the bidirectional feature synthesis module as input and decompose it into two independent components: the low-level feature retention component and the high-level constraint injection component; The bidirectional strength parameterization submodule is configured to map the two independent components after decomposition into connection strength parameters of the bottom connection strength and the high-level connection strength respectively; The weight attribute synthesis independently combines the two types of connection strength parameters into a bidirectional weight attribute, whose structure contains two independent dimensions: the weight value from the bottom layer to the new node and the weight value from the new node to the upper layer; The rule topology submodule is configured to construct two types of connection edges for new nodes in the initial data network based on bidirectional weight attributes: bottom-level connection edges and high-level connection edges; the combination of the two types of edges constitutes the complete connection rules of the cross-layer compensation path.

5. The data acquisition device for engineering information management according to claim 1, characterized in that: Bidirectional strength parameterization submodule, including: The component value extraction unit is configured to take the two independent components of the feature distribution decoupling analysis submodule output, the low-level feature retention component and the high-level constraint injection component, as input and extract their quantized values respectively; the extraction rules are as follows: a strength mapping rule activation unit configured to input the extracted values into a preset strength conversion program to generate connection strength parameters; The parameter normalization encapsulation unit is configured to limit the two types of strength parameters within a preset range and perform normalization alignment so that: the bottom-level connection strength parameter represents the priority of detail feature transmission, and the high-level connection strength parameter represents the urgency of constraint correction.

6. The data acquisition device for engineering information management according to claim 1, characterized in that: Intensity mapping rule activation unit, including: The conversion rule loading subunit is configured to take the low-level feature preservation component value output by the component value extraction unit and the high-level constraint injection component value as input and activate a preset intensity conversion program; a dynamic range calibration subunit, configured to perform range constraints and behavior correction on the initial intensity parameters; The parameter semantic alignment subunit is configured to semantically associate the calibrated bottom-level connection strength parameters with the high-level connection strength parameters; the bottom-level connection strength parameters and the high-level connection strength parameters together constitute the final connection strength parameters.

7. The data acquisition device for engineering information management according to claim 1, characterized in that: Engineering data representation system, including: The positive and negative example construction module is configured to construct two types of positive and negative example pairs based on the cross-layer compensation path of the multi-layer data fusion network and the hierarchical progressive path of the initial data network; The hierarchical difference measurement module is configured to define hierarchical difference measurement rules based on positive example pairs and negative example pairs; perform fusion network difference measurement and initial network difference measurement; The dual-network collaborative optimization module is configured to input the results of the two types of fused network difference metrics and the initial network difference metrics into the loss function and perform simultaneous optimization: The optimization result aggregation module is configured to perform hierarchical feature superposition of the optimized multi-level data fusion network node representation and the initial data network node representation; and finally generate an engineering data representation that is both stable and adaptable.

8. The data acquisition device for engineering information management according to claim 7, characterized in that: Positive example pair generation in the positive and negative example construction module: In the multi-layer data fusion network, new nodes and bottom / high-level node pairs connected by the same cross-layer compensation path are selected; in the initial data network, node pairs whose KL divergence difference between adjacent layer nodes is lower than a dynamic threshold are selected; Negative example pair generation: In a multi-level data fusion network, randomly select a new node with no connection path and any high-level node pair; in the initial data network, select node pairs whose cross-layer KL divergence difference exceeds a dynamic threshold.

9. The data acquisition device for engineering information management according to claim 1, characterized in that: Also includes: The data network construction system is configured to obtain sensor data, environmental monitoring data and manual input data from the engineering site, build an initial data network, and calculate the feature weight distribution of each node.

10. The data acquisition device for engineering information management according to claim 1, wherein: Also includes: The engineering data representation system is configured to construct positive example pairs and negative example pairs in a multi-level data fusion network, aggregate node representations in the multi-level data fusion network, and obtain a final engineering data representation.

Images