Real-time data collection method for water transportation engineering supervision based on project analysis

Abstract

The application discloses a water transportation engineering supervision real-time data collection method based on project analysis and relates to the technical field of engineering data collection. The method comprises the following steps: collecting water transportation engineering supervision data and preprocessing; constructing a supervision heterogeneous event graph structure for expressing the relationship among engineering parts, construction procedures and water area disturbance; inputting an improved graph neural ordinary differential equation network for state coding, continuous evolution and event jump correction, generating a collection result through missing collection and candidate collection counterfactual deduction; performing information gain analysis on the collection result, screening and sorting candidate collection actions in combination with evidence chain closed state, generating a collection strategy; generating a supervision data stream according to the strategy for executing real-time collection, performing consistency checking and coverage updating, simultaneously updating a data set, an event graph and a network model, and generating a supervision collection result. The application realizes real-time and efficient collection and multi-source fusion analysis of water transportation engineering supervision data.

Description

Technical Field

[0001] This invention relates to the field of engineering data acquisition technology, and in particular to a real-time data acquisition method for waterway engineering supervision based on project analysis. Background Technology

[0002] In current waterway engineering supervision processes, data collection and status recording largely rely on manual inspections and decentralized information systems. Data sources include construction site records, monitoring equipment data, vessel and machinery positioning information, and aquatic environment observation information. Existing technologies typically acquire data through fixed-period or event-triggered data collection methods, and store and query the data using databases or simple association models to support project progress management and quality supervision.

[0003] In the aforementioned existing technologies, there is a lack of a unified structured representation for supervision data from different sources. The correlation between information on project locations, construction procedures, and the aquatic environment is weak, making it difficult to conduct collaborative analysis of multi-source data within a unified model. Data processing typically employs static rules or simple statistical methods to assess the supervision status, which fails to depict the dynamic coupling between the construction process and aquatic disturbances, and also fails to reflect the uncertain changes in supervision conclusions under different data integrity conditions.

[0004] Existing technologies largely rely on preset rules or experience-based threshold triggering mechanisms for data acquisition strategies, lacking mechanisms to dynamically optimize acquisition behavior based on state evolution results. This leads to a low degree of matching between acquisition behavior and actual engineering conditions. During data updates, multi-source data is typically written directly to the database incrementally without maintaining consistency of graph-structured relationships or recalculating state change propagation paths. This results in lag and local distortion in subsequent analysis results, affecting the real-time performance and accuracy of supervisory decisions.

[0005] Therefore, how to provide a real-time data acquisition method for waterway engineering supervision based on project analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a real-time data acquisition method for waterway engineering supervision based on project analysis. This invention utilizes a heterogeneous event graph structure and a time-series graph-based constant differential equation model to collaboratively model the engineering status, waterway disturbance status, and uncertain status of supervision conclusions. It combines incomplete and candidate data acquisition counterfactual deduction to achieve dynamic optimization of the acquisition strategy, thereby possessing the advantages of strong data correlation, high adaptability of acquisition decisions, and high accuracy of supervision status analysis.

[0007] The real-time data acquisition method for waterway engineering supervision based on project analysis according to an embodiment of the present invention includes the following steps:

[0008] Collect waterway engineering supervision data, perform data preprocessing, and generate a preprocessed dataset;

[0009] Construct a heterogeneous event graph structure for supervision based on the preprocessed dataset;

[0010] The heterogeneous event graph structure of the supervision is input into the improved graph neural network of ordinary differential equations, and comprehensive state coding, continuous state evolution, event jump correction, missing collection counterfactual inference and candidate collection counterfactual inference are performed to generate collection inference results.

[0011] Perform accompanying sensitivity information gain analysis on the collected simulation results, and sort and filter candidate collection actions in combination with the evidence chain closure state in the heterogeneous event diagram structure of the supervision, and generate collection value ranking results;

[0012] A proactive data collection strategy for supervision is generated based on the ranking of collected data by value.

[0013] Based on the proactive data collection strategy for waterway engineering supervision, real-time data collection is performed to generate a real-time data stream for supervision.

[0014] The system performs consistency checks and coverage updates on the real-time data stream of the supervision project. Based on the check and update results, it updates the preprocessed dataset, the heterogeneous event graph structure of the supervision project, and the improved graph neural network of frequent differential equations to generate real-time data acquisition results for waterway engineering supervision.

[0015] Optionally, the process of generating the preprocessed dataset includes:

[0016] The supervision data is identified by source and aligned by time. Data is matched according to the project location and construction procedure. Missing, abnormal and invalid data are removed. Fields of valid data are standardized to generate normalized supervision data.

[0017] Based on the normalized supervision data, data on tide level, water flow, wind and waves, ship and engine offset, and water depth deviation are extracted to generate water disturbance identification data. The normalized supervision data and water disturbance identification data are then correlated according to the project location and collection time to generate a preprocessed dataset.

[0018] Optionally, the process of constructing the heterogeneous event graph structure for supervision includes:

[0019] Establish corresponding nodes based on the project location, construction sequence, supervision matters, source identification, and water area disturbance identification data;

[0020] Based on the correspondence between engineering parts and construction procedures, the temporal relationship between construction procedures, the correlation between supervision matters, and the correlation between water disturbance identification data and construction activities, establish related edges;

[0021] By combining the nodes and associated edges, a heterogeneous event graph structure for supervision is generated.

[0022] Optionally, the structure of the improved graphical ordinary differential equation network includes:

[0023] The input layer is used to receive the heterogeneous event graph structure of the supervision as structural input.

[0024] The state coding layer is used to encode the node features and edge relationships in the heterogeneous event graph structure of the supervision system, and generate coded state representations.

[0025] The continuous evolution layer is used to perform continuous state evolution on the encoded state representation based on the improved graph-based ordinary differential equation to generate the current state evolution result;

[0026] The event transition correction layer is used to perform discontinuous state correction on the current state evolution result when the preset supervision event triggering conditions are met, and generate the corrected supervision comprehensive state hidden variables.

[0027] The missing data collection counterfactual deduction branch is used to infer the state changes under the condition that no new data collection actions were performed based on the modified hidden variables of the comprehensive supervision state, and generate the missing data collection counterfactual evolution results;

[0028] The candidate acquisition counterfactual inference branch is used to infer the state changes under different candidate acquisition actions based on the corrected state results, and generate candidate acquisition counterfactual evolution results;

[0029] The accompanying sensitivity information gain analysis layer is used to compare and calculate the evolution results of the missing acquisition counterfactual and the candidate acquisition counterfactual, generate the acquisition information gain value corresponding to each candidate acquisition action, and output the acquisition value ranking result.

[0030] The output layer is used to generate priority output results for candidate collection actions based on the collection value ranking results.

[0031] Optionally, the process of synthesizing state coding and event transition correction includes:

[0032] Based on the heterogeneous event graph structure of the supervision, the heterogeneous nodes of the supervision are fused and encoded, and the associated edges in the heterogeneous event graph structure are combined for joint modeling to generate the hidden variables of the comprehensive state of supervision.

[0033] The hidden variables of the overall supervisory status undergo continuous state evolution;

[0034] During continuous state evolution, an event transition correction is executed when the preset supervision event triggering conditions are met;

[0035] When a supervision event trigger condition occurs, the corresponding state component in the supervision comprehensive state latent variable is updated discontinuously according to the changes in the corresponding node attributes and edge relationships, generating the corrected supervision comprehensive state latent variable.

[0036] Optionally, the process of counterfactual inference for missing samples and counterfactual inference for candidate samples includes:

[0037] Using the modified hidden variables of the comprehensive supervision status as the initial state node, the corresponding state transition function is called step by step according to the order of the supervision collection time slices in the heterogeneous event graph structure of the supervision to continuously update the state and construct the supervision status baseline path.

[0038] Without introducing new data collection actions, the baseline path of the supervision status is continuously deduced to generate the counterfactual evolution results of missing data collection.

[0039] For different candidate collection actions, a collection condition path is constructed, and the state of the collection condition path is deduced to generate candidate collection counterfactual evolution results;

[0040] The counterfactual evolution results of missing samples are compared and calculated with the counterfactual evolution results of candidate samples to generate the sample inference results.

[0041] Optionally, the process of performing the accompanying sensitivity information gain analysis includes:

[0042] Based on the data collection and simulation results, a time slice alignment sequence for the supervision data collection is constructed, and time alignment results are generated.

[0043] Node-by-node difference calculations are performed on the engineering status, data status, water disturbance status, and uncertain status of the supervision conclusion corresponding to the time slice collected by the supervision in the time alignment results to generate a state change difference sequence.

[0044] Based on the state change difference sequence, the convergence degree of the uncertain state of the supervision conclusion corresponding to the candidate collection action is calculated cumulatively over time slice to generate an uncertain state convergence sequence.

[0045] Weighted aggregation calculations are performed on the converged sequence of uncertain states to generate information gain values ​​corresponding to candidate acquisition actions. The information gain values ​​of each candidate acquisition action are then sorted to generate a ranking result of acquisition value.

[0046] Optionally, the process of generating the proactive data collection strategy for supervision includes:

[0047] Candidate collection actions are screened, and candidate collection actions whose evidence chains for corresponding supervision matters have been closed are removed. Based on the engineering parts nodes, construction process nodes, and water disturbance nodes corresponding to the retained candidate collection actions, a collection target set is generated.

[0048] Match the target data collection point with the corresponding data source node and generate the data source matching result;

[0049] Based on the matching results of the acquisition sources and the sorting results of the acquisition information gain, the candidate acquisition actions are prioritized and acquisition priority results are generated.

[0050] A proactive data collection strategy for supervision is generated based on the collection priority results.

[0051] Optionally, the process of performing consistency verification and coverage update includes:

[0052] Based on the proactive data collection strategy for supervision, the data collection operation is performed to obtain newly added real-time supervision data and form corresponding data update records;

[0053] Match the data update records with the existing data in the heterogeneous event graph structure of the supervision, incrementally update the attributes and relationships of the successfully matched nodes, and establish corresponding relationships for the unmatched data and write them into the heterogeneous event graph structure of the supervision.

[0054] Based on the updated node attributes and relationships, the feature representation of the affected nodes is recalculated, and the update results are propagated along the associated paths to complete the state synchronization update of the heterogeneous event graph structure of the supervision.

[0055] The synchronously updated graph structure of the heterogeneous events of the supervision is input into the graph neural network of constant differential equations, and the hidden variables of the comprehensive state of the supervision are solved again to obtain the updated evolution results of the supervision state.

[0056] Re-execute the counterfactual inference of missing samples and the counterfactual inference of candidate samples, and calculate the change in information gain;

[0057] The candidate data collection object ranking results are updated based on the recalculated information gain change, and the next round of proactive data collection strategy for supervision is generated.

[0058] Optionally, the process of re-executing the counterfactual inference of missing samples and the counterfactual inference of candidate samples includes:

[0059] Based on the updated heterogeneous event graph structure of the supervision and the updated evolution results of the supervision status, the baseline path of the supervision status is reconstructed. Based on the baseline path of the supervision status, the counterfactual inference of missing data collection and the counterfactual inference of candidate data collection are re-executed, and the information gain change of each candidate data collection object is updated.

[0060] The beneficial effects of this invention are:

[0061] (1) By constructing a heterogeneous event graph structure for supervision and introducing a time-series graph-based frequent differential equation model, the status of engineering parts, construction procedures and water disturbance are uniformly expressed and continuously evolved, realizing the structured fusion and dynamic correlation analysis of multi-source supervision data, and improving the consistency and temporal continuity of supervision status expression.

[0062] (2) By introducing the counterfactual inference mechanism of missing data collection and the counterfactual inference mechanism of candidate data collection, the evolution results of the supervision status under different data collection strategies are compared and analyzed, so as to realize the quantitative evaluation of the impact of data collection behavior on the uncertainty of supervision conclusions and improve the rationality and adaptive optimization capability of supervision data collection strategy.

[0063] (3) By using the acquisition decision mechanism based on the accompanying sensitivity information gain analysis and the evidence chain closure constraint, the candidate acquisition actions are sorted by information gain and constrained for selection, so as to realize the dynamic generation and optimization of the active acquisition strategy of supervision data, and improve the accuracy of supervision data acquisition and the reliability of engineering status identification. Attached Figure Description

[0064] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0065] Figure 1 This is a flowchart of the real-time data acquisition method for waterway engineering supervision based on project analysis proposed in this invention;

[0066] Figure 2 This is a schematic diagram of the supervision state evolution and event jump correction structure of the real-time data acquisition method for waterway engineering supervision based on project analysis proposed in this invention.

[0067] Figure 3 This invention presents a diagram illustrating the process of constructing a heterogeneous event graph for waterway engineering supervision using a real-time data acquisition method based on project analysis.

[0068] Figure 4 This is an improved graphical ordinary differential equation network structure diagram of the real-time data acquisition method for waterway engineering supervision based on project analysis proposed in this invention. Detailed Implementation

[0069] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0070] refer to Figures 1-4 The real-time data acquisition method for waterway engineering supervision based on project analysis includes the following steps:

[0071] Collect waterway engineering supervision data, perform data preprocessing, and generate a preprocessed dataset;

[0072] Construct a heterogeneous event graph structure for supervision based on the preprocessed dataset;

[0073] The heterogeneous event graph structure of the supervision is input into the improved graph neural network of ordinary differential equations, and comprehensive state coding, continuous state evolution, event jump correction, missing collection counterfactual inference and candidate collection counterfactual inference are performed to generate collection inference results.

[0074] Perform accompanying sensitivity information gain analysis on the collected simulation results, and sort and filter candidate collection actions in combination with the evidence chain closure state in the heterogeneous event diagram structure of the supervision, and generate collection value ranking results;

[0075] A proactive data collection strategy for supervision is generated based on the ranking of collected data by value.

[0076] Based on the proactive data collection strategy for waterway engineering supervision, real-time data collection is performed to generate a real-time data stream for supervision.

[0077] The system performs consistency checks and coverage updates on the real-time data stream of the supervision project. Based on the check and update results, it updates the preprocessed dataset, the heterogeneous event graph structure of the supervision project, and the improved graph neural network of frequent differential equations to generate real-time data acquisition results for waterway engineering supervision.

[0078] In this embodiment, the process of generating the preprocessed dataset includes:

[0079] The supervision data is identified by source and aligned by time. Data is matched according to the project location and construction procedure. Missing, abnormal and invalid data are removed. Fields of valid data are standardized to generate normalized supervision data.

[0080] Supervision data comprises on-site collected data and supervision business data generated during the supervision of waterway engineering projects. Specifically, it consists of construction progress records, quality inspection records, safety inspection records, measurement verification records, on-site supervision records, equipment operation records, ship and machinery positioning records, image acquisition records, and waterway condition observation records. The source identifier is used to record the data source type, acquisition terminal number, acquisition personnel number, inspection record number, and supervision task number corresponding to each piece of supervision data. Time alignment is used to correct the acquisition time of supervision data from different sources according to a unified supervision acquisition timeline, and to write data within the same supervision acquisition cycle into the same time slice, so that data from different sources can be matched under the same engineering condition.

[0081] Data matching is based on the project location code, construction procedure code, supervision item number, spatial coordinate range, and supervision data collection time slice, binding supervision data to the corresponding project location and construction procedure. Data that meets the conditions of consistency in project location, construction procedure, spatial coordinates, and collection time is considered a successful match. When multiple matching results exist, a unique result is determined based on spatial coordinate distance, construction procedure priority, and proximity of collection time. Valid data is data with complete source identifier, valid collection time, successful matching, and fields and values ​​that meet preset requirements. Invalid data is data with missing source identifier, invalid time, failed matching, missing key fields, abnormal values, or data with irresolvable conflicts.

[0082] Based on the normalized supervision data, data on tide level, water flow, wind and waves, ship and engine offset, and water depth deviation are extracted to generate water disturbance identification data. The normalized supervision data and water disturbance identification data are then correlated according to the project location and collection time to generate a preprocessed dataset.

[0083] Field standardization involves converting fields from different sources in valid data into fields with unified field names, unified units of measurement, unified time formats, unified spatial coordinate formats, and unified engineering part coding formats. For data from different sources under the same supervision item, the standardized data is aggregated according to the supervision item number, engineering part code, construction procedure code, and collection time slice to form unified supervision data that can be directly called by the heterogeneous supervision event diagram.

[0084] The unified supervision data has completed source identification, field standardization, time alignment, and engineering location matching. Water area condition observation records, vessel and machinery positioning records, measurement verification records, and construction equipment operation records all have a unified field structure within the unified supervision data. Tide data is extracted from the water level field in the water area condition observation records; water flow data is extracted from the velocity and direction fields; wind and wave data is extracted from the wind speed, wind direction, and wave height fields; vessel and machinery offset data is generated by comparing the real-time coordinates in the vessel and machinery positioning records with the planned operation coordinates of the corresponding construction process; and water depth deviation data is generated by comparing the measured water depth in the measurement verification records with the design water depth of the corresponding engineering location. Through the above field extraction and difference comparison, the unified supervision data can generate water area disturbance identification data corresponding to the engineering location and collection time.

[0085] The association process uses the project location code and the supervision collection time slice as a joint index to write the unified supervision data and water disturbance identification data within the same project location code and the same supervision collection time slice into the same data record. When the collection time of the water disturbance identification data is not completely consistent with the collection time of the unified supervision data, the data within the same supervision collection time slice and with the closest collection time is selected for association. When there are multiple water disturbance identification data for a project location within the same supervision collection time slice, the associated data is determined according to the priority of the data source and the proximity of the collection time. After the association is completed, each preprocessed data simultaneously contains the supervision business field, the water disturbance field, the project location field, and the collection time field, forming a preprocessed dataset.

[0086] In this embodiment, the process of constructing the heterogeneous event graph structure for supervision includes:

[0087] Establish corresponding nodes based on the project location, construction sequence, supervision matters, source identification, and water area disturbance identification data;

[0088] Based on the correspondence between engineering parts and construction procedures, the temporal relationship between construction procedures, the correlation between supervision matters, and the correlation between water disturbance identification data and construction activities, establish related edges;

[0089] By combining the nodes and associated edges, a heterogeneous event graph structure for supervision is generated.

[0090] When constructing a heterogeneous event graph for waterway engineering supervision based on the preprocessed dataset, the field text of the engineering location, construction procedure, supervision item, source identifier, and water disturbance identification data is not directly used as nodes. Instead, the normalized supervision data and water disturbance identification data in the preprocessed dataset are deduplicated and aggregated to generate node identifiers for the aggregated objects.

[0091] Specifically, the preprocessed dataset reads the corresponding engineering location, construction procedure, supervision item, source identifier, water area disturbance identification data, and collection time for each data record. For engineering locations, deduplication and aggregation are performed, grouping data with the same engineering location code or name and spatial range into one engineering location node, and writing the engineering location code, spatial coordinate range, and associated engineering area to the node. For construction procedures, deduplication and aggregation are performed, grouping data with the same construction procedure code or name under the same engineering location into one construction procedure node, and writing the construction procedure code, construction sequence, and corresponding engineering location to the node. For supervision items, deduplication and aggregation are performed, grouping data corresponding to the same engineering location, construction procedure, and supervision item into one supervision item node, and writing the supervision item number, supervision item type, collection time, and evidence chain closure status to the node. For source identifiers, deduplication and aggregation are performed, grouping data with the same collection source type, collection terminal number, or record source number into one collection source node, and writing the source type, source number, and source credibility status to the node. The water disturbance identification data is deduplicated and aggregated. Data corresponding to tidal changes, water flow changes, wind and wave changes, ship and engine offsets, and water depth deviations at the same engineering location and within the same collection time are merged into water disturbance nodes. The disturbance type, disturbance value, disturbance direction, and corresponding collection time are written to the water disturbance nodes.

[0092] After establishing nodes, association edges are created based on the correspondence between data records in the preprocessed dataset. If the same data record contains both an engineering location and a construction procedure, a procedure affiliation edge is created between the corresponding engineering location node and the corresponding construction procedure node to indicate that the construction procedure occurred in the corresponding engineering location. If two construction procedures under the same engineering location have a sequential relationship in terms of collection time or construction sequence, a procedure dependency edge is created between the preceding construction procedure node and the following construction procedure node to indicate the sequential connection relationship of the construction procedures. If two engineering locations are adjacent in spatial coordinate range and have related construction activities within the same collection time, a spatial adjacency edge is created between the two engineering location nodes to indicate the spatial influence relationship between adjacent engineering locations. If the same data record contains both a supervision item and a corresponding engineering location or construction procedure, a supervision constraint edge is created between the supervision item node and the corresponding engineering location node and the corresponding construction procedure node to indicate the constraint relationship of the supervision item on the engineering location and construction procedure. If the same data record contains a source identifier and a supervision item, a source binding edge is created between the collection source node and the supervision item node to indicate the collection source of the data corresponding to the supervision item. If the water disturbance identification data corresponds to a certain engineering part or a certain construction process in terms of collection time, then a disturbance association edge is established between the water disturbance node and the corresponding engineering part node and the corresponding construction process node to represent the influence relationship of the water disturbance on the corresponding engineering part and construction process.

[0093] The evidence chain closure status is generated based on the completeness of construction records, supervision records, testing records, image records, and review records under the same supervision item, and written to the corresponding supervision item node. When all the above records for the same supervision item exist and the collection time falls within the same supervision collection cycle, the corresponding supervision item node is marked as having a closed evidence chain; when any record is missing or there are mismatched engineering parts, construction procedures, or collection times among the records, the corresponding supervision item node is marked as having an incomplete evidence chain. The evidence chain closure status participates in the subsequent candidate collection action sorting and screening.

[0094] Through the aforementioned node establishment and edge construction operations, the heterogeneous event graph structure for supervision consists of nodes representing engineering locations, construction procedures, supervision items, data collection sources, and waterway disturbances, as well as edges representing procedure affiliation, procedure dependence, spatial adjacency, supervision constraints, source binding, and disturbance association. Each edge has a clearly defined node source, node destination, and data collection time, which can characterize the relationships between waterway engineering supervision data and engineering locations, construction processes, supervision items, data collection sources, and waterway disturbances.

[0095] In this embodiment, the structure of the improved graphical ordinary differential equation network includes:

[0096] The input layer is used to receive the heterogeneous event graph structure of the supervision as structural input.

[0097] The state coding layer is used to encode the node features and edge relationships in the heterogeneous event graph structure of the supervision system, and generate coded state representations.

[0098] The continuous evolution layer is used to perform continuous state evolution on the encoded state representation based on the improved graph-based ordinary differential equation to generate the current state evolution result;

[0099] The event transition correction layer is used to perform discontinuous state correction on the current state evolution result when the preset supervision event triggering conditions are met, and generate the corrected supervision comprehensive state hidden variables.

[0100] The missing data collection counterfactual deduction branch is used to infer the state changes under the condition that no new data collection actions were performed based on the modified hidden variables of the comprehensive supervision state, and generate the missing data collection counterfactual evolution results;

[0101] The candidate acquisition counterfactual inference branch is used to infer the state changes under different candidate acquisition actions based on the corrected state results, and generate candidate acquisition counterfactual evolution results;

[0102] The accompanying sensitivity information gain analysis layer is used to compare and calculate the evolution results of the missing acquisition counterfactual and the candidate acquisition counterfactual, generate the acquisition information gain value corresponding to each candidate acquisition action, and output the acquisition value ranking result.

[0103] The output layer is used to generate priority output results for candidate collection actions based on the collection value ranking results.

[0104] In this embodiment, the process of integrating state coding and event transition correction includes:

[0105] Based on the heterogeneous event graph structure of the supervision, the heterogeneous nodes of the supervision are fused and encoded, and the associated edges in the heterogeneous event graph structure are combined for joint modeling to generate the hidden variables of the comprehensive state of supervision.

[0106] Fusion coding refers to mapping the node attribute vectors corresponding to engineering part nodes, construction process nodes, supervision item nodes, data collection source nodes, and water disturbance nodes to a unified dimension, performing vector splicing and weighted aggregation in the same feature space, and introducing the structural weight information of their corresponding associated edges to modulate the connection strength of different nodes in the graph structure, thereby forming a unified state representation with structure perception capabilities.

[0107] Joint modeling refers to constructing a unified temporal evolution topology based on the latent variables of the integrated supervision status after fusion encoding, according to the node types and edge types in the heterogeneous event graph structure of supervision. It maps the dependencies between different types of nodes into temporal propagation paths, enabling the collaborative evolution constraints of project status, data status, water disturbance status, and uncertain status of supervision conclusions in a unified graph structure. Uncertain status of supervision conclusions refers to the state variables that quantify the reliability of supervision conclusions based on the evidence integrity and multi-source data consistency of supervision item nodes.

[0108] The implicit variables of the overall supervision status include project status, data status, water area disturbance status, and uncertainty of supervision conclusions. Project status is generated by the construction progress relationship represented by project location nodes, construction procedure nodes, and their associated edges. Data status is generated by the data integrity and consistency represented by data collection source nodes, supervision item nodes, and source binding edges. Water area disturbance status is generated by water area disturbance nodes and disturbance-related edges. Uncertainty of supervision conclusions is generated by the evidence association relationship represented by supervision item nodes and item-related edges.

[0109] The hidden variables of the comprehensive state of the supervision are continuously evolved. Specifically, the continuous state evolution is to model the hidden variables of the comprehensive state of the supervision in the continuous time domain based on the time sequence graph constant differential equation, establish the state differential update relationship between adjacent collection time slices, and perform propagation calculation on the node state changes according to the adjacency relationship in the heterogeneous event graph structure of the supervision, so as to obtain the state evolution result sequence that changes continuously with time.

[0110] During the continuous state evolution process, an event transition correction is executed when the preset supervision event triggering conditions are met;

[0111] Preset supervision event triggering conditions refer to the pre-defined supervision event identification rules used to determine whether to execute event transition correction. These rules include: process switching events triggered when the current process status of a construction process node changes from the previous process to the next process; concealed works inspection events triggered when a concealed works project corresponding to an engineering part node is completed and an acceptance application is submitted; inspection indicator exceeding limits events triggered when the quality inspection indicators, dimensional inspection indicators, strength inspection indicators, flatness indicators, compaction indicators, or material performance indicators associated with a supervision item node exceed preset control thresholds; and the ship / machinery positioning trajectory associated with the data collection source node and the boundary of the design and construction area, and the planned construction route. The following events trigger the event: a ship trajectory deviation event when the deviation distance between the control ranges exceeds a preset deviation threshold; a water depth remeasurement deviation event when the difference between the measured water depth data associated with the water disturbance node and the design water depth data, historical remeasurement data, or allowable deviation range exceeds a preset limit; a supervision record missing event when the inspection records, on-site records, inspection records, acceptance records, image data, or sensor data associated with the supervision item node are not uploaded, not updated, or have missing key fields within a preset time window; and an event jump trigger condition is determined to be met when any of the above events occurs in the supervision heterogeneous event diagram structure, or when multiple events occur simultaneously and meet the corresponding judgment rules.

[0112] When a supervision event trigger condition occurs, the corresponding state component in the supervision comprehensive state latent variable is updated discontinuously according to the changes in the corresponding node attributes and edge relationships, and the corrected supervision comprehensive state latent variable is generated.

[0113] Discontinuous update refers to the local reset or incremental correction of the node state vector corresponding to the changed engineering state, data state, water disturbance state, or uncertain state of supervision conclusion after the event jump is triggered. The edge propagation weight between the node and its adjacent nodes is recalculated to correct the breakpoint of the original continuous evolution state. This allows the latent variables of the corrected supervision comprehensive state to satisfy the constraint relationship of the current supervision heterogeneous event graph structure again.

[0114] For example, during the construction supervision of a waterway wharf, the system continuously collects construction site data, ship and machinery positioning data, water environment monitoring data, and supervision record data, and aggregates the data according to the preset supervision collection time slots.

[0115] Within a certain time frame of the supervision data collection, the engineering location node corresponds to "No. 5 caisson area", the construction procedure node corresponds to "caisson installation operation", the supervision item node corresponds to "caisson positioning deviation detection", the data collection source nodes include "GNSS positioning terminal", "supervision mobile terminal" and "water level sensor", and the water disturbance node corresponds to "local tide level rise and flow velocity enhancement status".

[0116] The system performs fusion encoding on the above node data, maps the attribute vectors of engineering part nodes, construction process nodes, supervision item nodes, data collection source nodes, and water disturbance nodes to a unified feature space, and performs weighted aggregation by combining the structural weights of process dependency edges, process affiliation edges, source binding edges, item association edges, and disturbance association edges, thereby forming a comprehensive supervision status latent variable that can simultaneously represent the construction status, supervision status, and water environment status.

[0117] The system uses a heterogeneous event graph structure for supervision to jointly model the aforementioned hidden variables of the comprehensive state, transforming the dependencies between different nodes into time-series propagation paths, so that the caisson installation operation status, positioning detection results, and tide disturbance status can be propagated collaboratively in the same graph structure.

[0118] During the continuous state evolution process, the system continuously updates the hidden variables of the comprehensive state of the supervision based on the time sequence graph constant differential equation, and generates a sequence of state evolution results that change over time, which is used to characterize the coupling changes of structural position offset trend and environmental disturbance during the caisson installation process.

[0119] When the subsequent acquisition time slice detects that the caisson installation operation has switched from the "positioning stage" to the "calibration and adjustment stage", a process switching event is triggered; at the same time, the GNSS detection results show that the actual position of the caisson deviates from the design coordinates by more than a preset threshold, triggering a detection index exceeding limit event; and the water level sensor shows that the tide level is rising rapidly beyond the allowable range, triggering a water depth remeasurement deviation event.

[0120] When the above events occur, the system performs an event jump correction operation, which updates the corresponding engineering state components and water disturbance state components discontinuously. For example, it performs a partial reset correction on the caisson installation and positioning state, and an incremental correction on the tide disturbance state. It recalculates the association propagation weight between the caisson and the construction process nodes and supervision items nodes, so that the corrected comprehensive supervision state can once again meet the current event constraints.

[0121] The modified hidden variables of the overall state of the supervision are input again into the graph-based constant differential equation network to continue the state evolution.

[0122] In this embodiment, the processes of counterfactual deduction for missing data collection and counterfactual deduction for candidate data collection include:

[0123] Using the modified hidden variables of the comprehensive supervision status as the initial state node, the corresponding state transition function is called step by step according to the order of the supervision collection time slices in the heterogeneous event graph structure of the supervision to continuously update the state and construct the supervision status baseline path.

[0124] Continuous state update refers to the recursive calculation of the project state, data state, water disturbance state and uncertain state of supervision conclusions based on the relationship between the hidden variables of the current comprehensive supervision state and the nodes in the heterogeneous event graph structure of supervision, according to the state dependency relationship between adjacent supervision collection time slices. This allows the state variables to be gradually transmitted and updated in the time dimension, thereby forming a state sequence that changes continuously in time slices.

[0125] Without introducing new data collection actions, the baseline path of the supervision status is continuously deduced to generate the counterfactual evolution results of missing data collection.

[0126] Continuous deduction of the baseline path of the supervision status refers to, while keeping the structure of the heterogeneous event graph of supervision and its node relationships unchanged, relying only on the time propagation results of the existing hidden variables of the comprehensive supervision status, and performing the state propagation calculation step by step according to the time slice sequence of the supervision collection, so that the system can continue to generate the state change trajectory of subsequent time slices without introducing external collection information increments.

[0127] For different candidate collection actions, a collection condition path is constructed, and the state of the collection condition path is deduced to generate candidate collection counterfactual evolution results;

[0128] The data collection condition path refers to the state propagation path formed in the heterogeneous event graph structure of the supervision, which takes the target engineering part node, construction process node, water disturbance node or supervision item node corresponding to the candidate data collection action as the starting node of the path, expands the path according to the process dependency edge, process ownership edge, item association edge, disturbance association edge and source binding edge, and filters the expandable nodes in combination with the order of supervision data collection time slices.

[0129] The construction of the data collection path includes: determining the set of target nodes corresponding to the candidate data collection actions; using the set of target nodes as the initial path nodes; expanding the adjacency based on the associated edges in the heterogeneous event graph structure of the supervision; and filtering the expanded nodes according to the time slice order of the supervision data collection to form a time-progressive path sequence.

[0130] State extrapolation of the acquisition condition path refers to introducing the observation increment corresponding to the candidate acquisition action at the corresponding engineering part node, construction process node or water disturbance node under the condition that the structure of the heterogeneous event diagram of the supervision remains unchanged, and embedding the observation increment as the state update input into the time-series propagation process. Independent state evolution calculations are performed on the paths corresponding to different candidate acquisition actions. That is, a state propagation process is established separately for each candidate acquisition action, and the state value is updated node by node along the acquisition condition path under the condition that it is not affected by other candidate acquisition actions, thereby generating a state change sequence under the action of different acquisition actions.

[0131] The counterfactual evolution results of missing samples are compared and calculated with the counterfactual evolution results of candidate samples to generate the sample inference results;

[0132] Specifically, comparative calculation refers to aligning the counterfactual evolution results of missing data collection with the counterfactual evolution results of each candidate data collection based on the same time slice of the supervision data collection. The calculation of differences in the amount of deviation in engineering status, the amount of change in data status integrity, the amount of change in water disturbance response, and the degree of convergence of the uncertain state of the supervision conclusion is performed to obtain comparative results on the improvement effect of different candidate data collection actions on the supervision status.

[0133] In this embodiment, the accompanying sensitivity information gain analysis is based on the present invention, which introduces a sensitivity propagation calculation mechanism for changes in the structural state of the heterogeneous event graph of the supervision during the counterfactual deduction of missing collection and the counterfactual deduction of candidate collection. By performing an accompanying tracking analysis on the degree of influence of the state evolution path under different collection actions, and combining the difference in uncertainty changes under missing collection and collection conditions, the contribution of each candidate collection action to the convergence of uncertainty of the supervision conclusion is quantified, forming an information gain evaluation method for collection optimization decision-making.

[0134] Specifically, the sensitivity propagation calculation mechanism refers to the process of propagating the influence of state changes of each node in the heterogeneous event graph structure of supervision on the final supervision conclusion output by layer-by-layer propagation calculation based on the state evolution relationship of graph god ordinary differential equation network. That is, by tracking the contribution changes of engineering part nodes, construction process nodes, supervision item nodes and water disturbance nodes in the state update process, the influence intensity of different candidate collection actions on the overall state evolution path is determined.

[0135] Uncertainty variation difference refers to the result of comparing and calculating the magnitude of change in the uncertainty state of the supervision conclusion between the counterfactual inference results of missing collection and the counterfactual inference results of candidate collection. Specifically, it is manifested by comparing the difference in the stability of the judgment of supervision matters, the level of data integrity, and the consistency of the interpretation of water disturbance under the same supervision collection time slice, so as to obtain the quantitative difference in the reduction or increase of uncertainty caused by different collection actions.

[0136] In this embodiment, the process of performing accompanying sensitivity information gain analysis includes:

[0137] Based on the counterfactual evolution results of missing data collection and candidate data collection in the data collection simulation results, a time slice alignment sequence for supervision data collection is constructed to generate time alignment results;

[0138] The alignment sequence of supervision collection time slices refers to the process of reordering and matching the engineering status, data status, water disturbance status, and uncertain status of supervision conclusions corresponding to the missing collection counterfactual evolution results and the candidate collection counterfactual evolution results according to the same time slice number, and supplementing the missing time slices by using the forward extension method, thereby forming an alignment sequence structure with a one-to-one correspondence in the time dimension.

[0139] Node-by-node difference calculations are performed on the engineering status, data status, water disturbance status, and uncertain status of the supervision conclusion corresponding to the time slice collected by the supervision in the time alignment results to generate a state change difference sequence.

[0140] Based on the state change difference sequence, the convergence degree of the uncertain state of the supervision conclusion corresponding to the candidate collection action is calculated cumulatively over time slice to generate an uncertain state convergence sequence.

[0141] The convergence degree of the uncertainty state of the supervision conclusion refers to the decrease in the uncertainty state of the supervision conclusion in the candidate collection counterfactual evolution result relative to the missing collection counterfactual evolution result within the same supervision collection time slice. It is used to characterize the ability of the candidate collection action to improve the certainty of the supervision conclusion.

[0142] Among them, the cumulative calculation of time slices refers to the process of stacking the convergence degree of the uncertain state corresponding to each time slice in the order of time slices collected by the supervisor, and recursively updating the change trend between consecutive time slices to form a cumulative convergence sequence that grows with time.

[0143] Weighted aggregation calculations are performed on the convergence sequence of uncertain states to generate information gain values ​​corresponding to candidate acquisition actions. Weighted aggregation calculations refer to assigning weights to the convergence degree of uncertain states in each time slice according to the importance of the acquisition time slices, and multiplying and summing the convergence values ​​of each time slice with their corresponding weights. The weights are determined based on the time slices where key engineering nodes are located, the key stages of construction procedures, and the degree of change in the intensity of water disturbance, resulting in information gain results that can characterize the overall contribution of different candidate acquisition actions. The information gain values ​​of each candidate acquisition action are then sorted to generate acquisition value ranking results.

[0144] In this embodiment, the process of generating a proactive data collection strategy for supervision includes:

[0145] Candidate collection actions are screened, and candidate collection actions whose evidence chains for corresponding supervision matters have been closed are removed. Based on the engineering parts nodes, construction process nodes, and water disturbance nodes corresponding to the retained candidate collection actions, a collection target set is generated.

[0146] Among them, generating a set of acquisition targets based on the engineering part nodes, construction process nodes and water disturbance nodes corresponding to the retained candidate acquisition actions means that the target nodes of the candidate acquisition actions in the heterogeneous event diagram structure of the supervision are aggregated and processed according to the correspondence between engineering part nodes and construction process nodes. Target nodes belonging to the same engineering part node or construction process node are merged into a unified acquisition target unit. At the same time, the disturbance type information and disturbance intensity information corresponding to the water disturbance nodes are retained to generate a structured set of acquisition targets.

[0147] Match the target data collection node with the corresponding data collection source node and generate the data collection source matching result;

[0148] Based on the matching results of the acquisition sources and the ranking results of the acquisition information gain, the candidate acquisition actions are prioritized and acquisition priority results are generated.

[0149] Matching the corresponding data source node refers to the process of searching for data source nodes that have source binding edges with each data source unit in the data collection target set, based on the engineering part nodes, construction process nodes, and water disturbance nodes corresponding to each data collection target unit, and then matching and filtering them according to the consistency of the spatial location, time slice, and data type between the data source identifier corresponding to the data collection source node and the data collection target unit.

[0150] Specifically, based on the engineering part nodes and construction process nodes in the target unit, a set of candidate acquisition source nodes with source binding edges is determined in the heterogeneous event graph structure of the supervision unit; then, for each acquisition source node in the candidate acquisition source node set, the time consistency relationship between its corresponding acquisition time slice and the supervision acquisition time slice of the target unit is calculated, and the spatial coverage of the acquisition source node is matched and verified with the spatial range of the engineering part node; among the acquisition source nodes that simultaneously satisfy the time slice consistency and spatial coverage consistency, the target acquisition source node is determined.

[0151] When multiple data collection source nodes meet the matching conditions at the same time, they are sorted and selected according to the priority of the source type. The priority of the source type is determined based on the data reliability identifier corresponding to the data collection source node in the heterogeneous event graph structure of the supervision, and a unique matching result of the data collection source node is generated.

[0152] A proactive data collection strategy for supervision is generated based on the collection priority results.

[0153] The proactive data collection strategy for supervision refers to the rule structure generated based on the ranking results of collected information gain, which guides the execution of supervision data collection. The rule structure is used to uniformly constrain and schedule the execution objects, execution content, execution time slices and execution sources of candidate collection actions, so as to determine the execution priority and execution order of different candidate collection actions.

[0154] For example, in the construction supervision scenario of a caisson wharf in a waterway project, after the system completes the information gain sorting within a certain supervision collection time slice, it obtains three candidate collection actions, whose corresponding collection information gain values ​​are A1>A2>A3 in order.

[0155] The proactive data collection strategy for the supervision data generated based on this sorting result is as follows:

[0156] Prioritize the data collection action for the "construction process node of the caisson area at Pier 3". The corresponding data to be collected are caisson positioning deviation data and construction progress data. The data collection time slice is the current data collection time slice of the supervisor. The data collection source is GNSS positioning terminal and supervisor mobile terminal.

[0157] The second-best execution is to collect data on "nearby water disturbance nodes", which includes tidal level change data and flow velocity change data. The data collection time slices are the current and the next supervisor's data collection time slices, and the data source is the water environment monitoring terminal.

[0158] The data collection action is performed for the "caisson acceptance supervision item node". The corresponding data collection content is structural deviation verification data and image recording data. The data collection time slice is the current supervision data collection time slice. The data collection source is the supervision image acquisition terminal and the manual verification terminal.

[0159] The above data collection actions are performed sequentially in descending order of information gain value. After each data collection time slice by the supervisor, the data collection results are updated, and the priority and execution order of subsequent data collection are adjusted.

[0160] In this embodiment, the process of performing consistency verification and coverage update includes:

[0161] Based on the proactive data collection strategy for supervision, the data collection operation is performed to obtain newly added real-time supervision data and form corresponding data update records;

[0162] Match the data update records with the existing data in the heterogeneous event graph structure of the supervision, incrementally update the attributes and relationships of the successfully matched nodes, and establish corresponding relationships for the unmatched data and write them into the heterogeneous event graph structure of the supervision.

[0163] Specifically, incremental update refers to the process of updating the attributes of only the engineering parts nodes, construction process nodes, supervision item nodes, and water disturbance nodes that have a matching relationship with the newly added real-time supervision data in the heterogeneous event graph structure of supervision. During the update process, the attributes of the unaffected nodes and their associated edge structure remain unchanged, and only the attribute values ​​of the nodes corresponding to the newly added data are superimposed or overwritten, thereby avoiding the reconstruction calculation of the entire graph structure and realizing the local dynamic evolution of the heterogeneous event graph structure of supervision.

[0164] Based on the updated node attributes and relationships, the feature representation of the affected nodes is recalculated, and the update results are propagated along the associated paths to complete the state synchronization update of the heterogeneous event graph structure of the supervision.

[0165] Recalculating the feature representation of the affected node refers to, after completing the node attribute update, taking the node with the changed attribute as the starting node, performing multi-hop neighborhood expansion along process dependency edge, process affiliation edge, source binding edge, event association edge and disturbance association edge, performing layer-by-layer feature recoding calculation on the nodes on the expansion path, and jointly aggregating the updated node attributes with the structural information of its neighboring nodes to generate the updated feature representation of the affected node. The update result is then propagated to adjacent nodes along the association path to realize the step-by-step diffusion process from local update to global state.

[0166] The synchronously updated graph structure of the heterogeneous events of the supervision is input into the graph neural network of constant differential equations, and the hidden variables of the comprehensive state of the supervision are solved again to obtain the updated evolution results of the supervision state.

[0167] Re-execute the counterfactual inference of missing samples and the counterfactual inference of candidate samples, and calculate the change in information gain;

[0168] The candidate data collection object ranking results are updated based on the recalculated information gain change, and the next round of proactive data collection strategy for supervision is generated.

[0169] In this embodiment, the process of re-executing the counterfactual inference of missing samples and the counterfactual inference of candidate samples includes:

[0170] Based on the updated heterogeneous event graph structure of the supervision and the updated evolution results of the supervision status, the baseline path of the supervision status is reconstructed. Based on the baseline path of the supervision status, the counterfactual inference of missing data collection and the counterfactual inference of candidate data collection are re-executed, and the information gain change of each candidate data collection object is updated.

[0171] Example: To verify the feasibility of this invention in practice, it was applied to a waterway engineering supervision scenario. Addressing issues such as fragmented information sources during construction, reliance on manual experience for data collection, and strong coupling between site conditions and changes in the aquatic environment, the method of this invention was applied to the continuous construction supervision data collection and analysis process. This scenario involves wharf caisson construction, monitoring of water tidal changes, and monitoring of construction machinery operation trajectories. Multiple types of supervision data are involved, including construction progress records, GNSS positioning data, water level and flow velocity monitoring data, and supervision inspection records. Traditionally, these data are stored separately in independent systems, lacking a unified structural representation. This forces supervisors to rely on manual summarization and experience-based judgment during status assessment, making it difficult to generate consistent dynamic analysis results.

[0172] In this embodiment, the aforementioned multi-source supervision data is first collected in a unified manner, and data preprocessing operations are performed. The preprocessing process includes time slice alignment, abnormal data removal, and field standardization, so that data from different sources are mapped to a unified supervision data expression form. Based on the preprocessed data, a heterogeneous supervision event graph structure is constructed, and the engineering parts, construction procedures, supervision matters, collection sources, and water disturbance information are respectively mapped as nodes in the graph structure. According to the process dependency relationship, the supervision record association relationship, and the water environment impact relationship, the association edges between nodes are established to form a structured supervision status expression model.

[0173] Based on the heterogeneous event graph structure of the supervision system, it is input into an improved graph-based constant differential equation network for state modeling. The system comprehensively encodes the node states and generates hidden variables for the comprehensive supervision state. It then performs state evolution calculations in the continuous time dimension, enabling the construction state, water disturbance state, and uncertain state of the supervision conclusions to be updated synchronously within the same evolutionary framework. When situations such as process switching or increased water disturbance occur during supervision, the system triggers an event jump correction mechanism to make discontinuous adjustments to the local states, thereby ensuring that the state evolution results reflect sudden changes on site.

[0174] Based on the state evolution results, the system constructs counterfactual inference paths for missing data collection and candidate data collection. The missing data collection counterfactual inference simulates the natural evolution trend of the supervision state without adding new data collection actions, while the candidate data collection counterfactual inference simulates the differences in the supervision state after different data collection actions are introduced. By aligning the two types of counterfactual paths with time slices, the system calculates the differences in project state deviation, data integrity changes, and uncertainty changes in supervision conclusions, forming a state change difference sequence. Furthermore, the system cumulatively calculates the degree of decrease in the uncertainty of supervision conclusions caused by different candidate data collection actions and performs weighted aggregation based on time slice weights to generate information gain evaluation results.

[0175] Based on the information gain result and the closed-loop state of the evidence chain, candidate collection actions are screened and sorted, and a proactive data collection strategy for supervision is generated. This strategy guides the collection execution order of different supervision objects under different time slices and different source conditions. After executing this strategy, the system automatically triggers the real-time data collection process for supervision, rewrites the collection results into the heterogeneous event graph structure of supervision, incrementally updates node attributes and associated edges, and recalculates the feature representations of affected nodes and propagates the updates along the graph structure to maintain dynamic consistency of the overall supervision state. The updated supervision state is then re-input into the graph-based frequent differential equation network for recalculation, thereby achieving closed-loop optimization between the collection strategy, state evolution, and counterfactual inference.

[0176] Through the above implementation methods, compared with traditional methods that rely on fixed-period data collection or manual experience-based judgment, this invention can adaptively adjust the data collection strategy during dynamic changes in the supervision status, improve the continuity and consistency of supervision data analysis under the coupled conditions of complex aquatic disturbance environment and construction status, and enhance the responsiveness of supervision conclusions to changes in multi-source data. Specific data are shown in Table 1 below.

[0177] Table 1: Comparison of Data Collection and Analysis Results in Waterway Engineering Supervision

[0178] Indicator Type Traditional supervision methods Method of the present invention Data acquisition delay (seconds) 45.8 12.6 Data integrity (%) 82.4 96.7 State recognition consistency rate (%) 79.1 95.3 Uncertainty Index of Supervision Conclusion 0.38 0.12 Abnormal state response time (seconds) 52.3 15.4 Multi-source data fusion error rate (%) 17.6 4.9 Number of times the data collection strategy is dynamically adjusted (per cycle) 1 6

[0179] Analysis of the data in Table 1 shows that the data acquisition and analysis capabilities for waterway engineering supervision significantly improve across several key indicators after the introduction of the method of this invention. Regarding data acquisition latency, it is reduced from 45.8 seconds using the traditional method to 12.6 seconds, indicating that the dynamic processing mechanism based on the heterogeneous event graph structure and time-series graph regular differential equations effectively shortens the processing time from data acquisition to structured representation. In terms of data integrity, the method of this invention improves from 82.4% to 96.7%, demonstrating that the fusion of multi-source supervision data and structured graph modeling reduces data loss and isolated information issues.

[0180] Regarding the consistency rate of state identification, it increased from 79.1% to 95.3%, reflecting that the continuous state evolution and event jump correction mechanism can enhance the adaptability of the supervision status to complex construction processes and changes in water disturbance, making the state judgment results between different supervision data sources more consistent. Regarding the uncertainty index of supervision conclusions, it decreased from 0.38 to 0.12, indicating that the counterfactual deduction and information gain analysis mechanism can effectively reduce the uncertainty of supervision judgments and improve the stability and reliability of supervision conclusions.

[0181] Regarding the response time to abnormal states, it decreased from 52.3 seconds to 15.4 seconds, indicating that event-driven state updates and local incremental calculation mechanisms can accelerate the identification and response speed of abnormal changes. Regarding the multi-source data fusion error rate, it decreased from 17.6% to 4.9%, demonstrating that unified graph structure modeling and association propagation mechanisms can improve the matching accuracy between different data sources. Regarding the number of times the acquisition strategy is dynamically adjusted, it increased from 1 time to 6 times, indicating that the active acquisition strategy based on information gain feedback can achieve higher-frequency adaptive optimization, enabling the supervision data acquisition process to have continuous iteration and dynamic optimization capabilities.

[0182] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A real-time data acquisition method for waterway engineering supervision based on project analysis, characterized in that, Includes the following steps: Collect waterway engineering supervision data, perform data preprocessing, and generate a preprocessed dataset; Construct a heterogeneous event graph structure for supervision based on the preprocessed dataset; The heterogeneous event graph structure of the supervision is input into the improved graph neural network of ordinary differential equations, and comprehensive state coding, continuous state evolution, event jump correction, missing collection counterfactual inference and candidate collection counterfactual inference are performed to generate collection inference results. Perform accompanying sensitivity information gain analysis on the collected simulation results, and sort and filter candidate collection actions in combination with the evidence chain closure state in the heterogeneous event diagram structure of the supervision, and generate collection value ranking results; A proactive data collection strategy for supervision is generated based on the ranking of collected data by value. Based on the proactive data collection strategy for waterway engineering supervision, real-time data collection is performed to generate a real-time data stream for supervision. The system performs consistency checks and coverage updates on the real-time data stream of the supervision project. Based on the check and update results, it updates the preprocessed dataset, the heterogeneous event graph structure of the supervision project, and the improved graph neural network of frequent differential equations to generate real-time data acquisition results for waterway engineering supervision.

2. The real-time data acquisition method for waterway engineering supervision based on project analysis according to claim 1, characterized in that, The process of generating the preprocessed dataset includes: The supervision data is identified by source and aligned by time. Data is matched according to the project location and construction procedure. Missing, abnormal and invalid data are removed. Fields of valid data are standardized to generate normalized supervision data. Based on the normalized supervision data, data on tide level, water flow, wind and waves, ship and engine offset, and water depth deviation are extracted to generate water disturbance identification data. The normalized supervision data and water disturbance identification data are then correlated according to the project location and collection time to generate a preprocessed dataset.

3. The real-time data acquisition method for waterway engineering supervision based on project analysis according to claim 2, characterized in that, The process of constructing the heterogeneous event graph structure for supervision includes: Establish corresponding nodes based on the project location, construction sequence, supervision matters, source identification, and water area disturbance identification data; Based on the correspondence between engineering parts and construction procedures, the temporal relationship between construction procedures, the correlation between supervision matters, and the correlation between water disturbance identification data and construction activities, establish related edges; By combining the nodes and associated edges, a heterogeneous event graph structure for supervision is generated.

4. The real-time data acquisition method for waterway engineering supervision based on project analysis according to claim 3, characterized in that, The structure of the improved graphical ordinary differential equation network includes: The input layer is used to receive the heterogeneous event graph structure of the supervision as structural input. The state coding layer is used to encode the node features and edge relationships in the heterogeneous event graph structure of the supervision system, and generate coded state representations. The continuous evolution layer is used to perform continuous state evolution on the encoded state representation based on the improved graph-based ordinary differential equation to generate the current state evolution result; The event transition correction layer is used to perform discontinuous state correction on the current state evolution result when the preset supervision event triggering conditions are met, and generate the corrected supervision comprehensive state hidden variables. The missing data collection counterfactual deduction branch is used to infer the state changes under the condition that no new data collection actions were performed based on the modified hidden variables of the comprehensive supervision state, and generate the missing data collection counterfactual evolution results; The candidate acquisition counterfactual inference branch is used to infer the state changes under different candidate acquisition actions based on the corrected state results, and generate candidate acquisition counterfactual evolution results; The accompanying sensitivity information gain analysis layer is used to compare and calculate the evolution results of the missing acquisition counterfactual and the candidate acquisition counterfactual, generate the acquisition information gain value corresponding to each candidate acquisition action, and output the acquisition value ranking result. The output layer is used to generate priority output results for candidate collection actions based on the collection value ranking results.

5. The real-time data acquisition method for waterway engineering supervision based on project analysis according to claim 4, characterized in that, The process of comprehensive state coding and event transition correction includes: Based on the heterogeneous event graph structure of the supervision, the heterogeneous nodes of the supervision are fused and encoded, and the associated edges in the heterogeneous event graph structure are combined for joint modeling to generate the hidden variables of the comprehensive state of supervision. The hidden variables of the overall supervisory status undergo continuous state evolution; During continuous state evolution, an event transition correction is executed when the preset supervision event triggering conditions are met; When a supervision event trigger condition occurs, the corresponding state component in the supervision comprehensive state latent variable is updated discontinuously according to the changes in the corresponding node attributes and edge relationships, generating the corrected supervision comprehensive state latent variable.

6. The real-time data acquisition method for waterway engineering supervision based on project analysis according to claim 5, characterized in that, The processes of the missing sample counterfactual deduction and the candidate sample counterfactual deduction include: Using the modified hidden variables of the comprehensive supervision status as the initial state node, the corresponding state transition function is called step by step according to the order of the supervision collection time slices in the heterogeneous event graph structure of the supervision to continuously update the state and construct the supervision status baseline path. Without introducing new data collection actions, the baseline path of the supervision status is continuously deduced to generate the counterfactual evolution results of missing data collection. For different candidate collection actions, a collection condition path is constructed, and the state of the collection condition path is deduced to generate candidate collection counterfactual evolution results; The counterfactual evolution results of missing samples are compared and calculated with the counterfactual evolution results of candidate samples to generate the sample inference results.

7. The real-time data acquisition method for waterway engineering supervision based on project analysis according to claim 6, characterized in that, The process of performing the accompanying sensitivity information gain analysis includes: Based on the data collection and simulation results, a time slice alignment sequence for the supervision data collection is constructed, and time alignment results are generated. Node-by-node difference calculations are performed on the engineering status, data status, water disturbance status, and uncertain status of the supervision conclusion corresponding to the time slice collected by the supervision in the time alignment results to generate a state change difference sequence. Based on the state change difference sequence, the convergence degree of the uncertain state of the supervision conclusion corresponding to the candidate collection action is calculated cumulatively over time slice to generate an uncertain state convergence sequence. Weighted aggregation calculations are performed on the converged sequence of uncertain states to generate information gain values ​​corresponding to candidate acquisition actions. The information gain values ​​of each candidate acquisition action are then sorted to generate a ranking result of acquisition value.

8. The real-time data acquisition method for waterway engineering supervision based on project analysis according to claim 7, characterized in that, The process of generating the proactive data collection strategy for supervision includes: Candidate collection actions are screened, and candidate collection actions whose evidence chains for corresponding supervision matters have been closed are removed. Based on the engineering parts nodes, construction process nodes, and water disturbance nodes corresponding to the retained candidate collection actions, a collection target set is generated. Match the target data collection point with the corresponding data source node and generate the data source matching result; Based on the matching results of the acquisition sources and the sorting results of the acquisition information gain, the candidate acquisition actions are prioritized and acquisition priority results are generated. A proactive data collection strategy for supervision is generated based on the collection priority results.

9. The real-time data acquisition method for waterway engineering supervision based on project analysis according to claim 8, characterized in that, The process of performing consistency verification and coverage update includes: Based on the proactive data collection strategy for supervision, the data collection operation is performed to obtain newly added real-time supervision data and form corresponding data update records; Match the data update records with the existing data in the heterogeneous event graph structure of the supervision, incrementally update the attributes and relationships of the successfully matched nodes, and establish corresponding relationships for the unmatched data and write them into the heterogeneous event graph structure of the supervision. Based on the updated node attributes and relationships, the feature representation of the affected nodes is recalculated, and the update results are propagated along the associated paths to complete the state synchronization update of the heterogeneous event graph structure of the supervision. The synchronously updated graph structure of the heterogeneous events of the supervision is input into the graph neural network of constant differential equations, and the hidden variables of the comprehensive state of the supervision are solved again to obtain the updated evolution results of the supervision state. Re-execute the counterfactual inference of missing samples and the counterfactual inference of candidate samples, and calculate the change in information gain; The candidate data collection object ranking results are updated based on the recalculated information gain change, and the next round of proactive data collection strategy for supervision is generated.

10. The real-time data acquisition method for waterway engineering supervision based on project analysis according to claim 9, characterized in that, The process of re-executing the counterfactual inference of missing samples and the counterfactual inference of candidate samples includes: Based on the updated heterogeneous event graph structure of the supervision and the updated evolution results of the supervision status, the baseline path of the supervision status is reconstructed. Based on the baseline path of the supervision status, the counterfactual inference of missing data collection and the counterfactual inference of candidate data collection are re-executed, and the information gain change of each candidate data collection object is updated.

Images