A method, system, terminal equipment, and media for constructing a railway engineering construction section-level platform based on temporal prediction and visual inspection networks.

Abstract

This invention discloses a method, system, terminal equipment, and medium for constructing a railway engineering construction section-level platform based on temporal prediction and visual inspection networks, belonging to the field of intelligent construction technology. The method includes: constructing a global perception layer to collect real-time field data through a multimodal sensor cluster; constructing a data processing layer to perform edge-side preprocessing and align the data with the BIM model in a spatiotemporal semantic manner to generate digital twin mapping data; constructing an intelligent engine layer to perform in-depth data mining using an AI model library; constructing a business application layer to transform the analysis results into business control commands to achieve closed-loop control; and constructing a user interaction layer to provide differentiated services. This invention, through a five-layer architecture design and deep AI empowerment, breaks down data silos in railway construction, achieving intelligent management and control of the entire process from perception and analysis to decision-making, significantly improving the precision of construction management and risk prevention capabilities.

Description

Technical Field

[0001] This invention relates to the field of railway construction platform construction technology, and in particular to a method, system, terminal equipment and medium for constructing a railway engineering construction section-level platform based on time-series prediction and visual inspection networks. Background Technology

[0002] Currently, the traditional railway engineering construction section management model mainly suffers from the following pain points: 1. Data collection is lagging and fragmented: It relies heavily on manual filling and paper records, resulting in low data accuracy and poor timeliness. Furthermore, data standards vary across different work areas and professions, making it difficult to achieve cross-professional data integration.

[0003] 2. Passive risk management: The discovery of delays, quality defects, and safety hazards mainly relies on manual inspections and experience-based judgment, lacking data-based predictive and early warning capabilities.

[0004] 3. Weak decision support: Although some information systems have been built, they are mostly simple information displays, lacking in-depth data mining and AI-assisted decision-making functions, and are unable to provide management with accurate resource scheduling and schedule optimization suggestions.

[0005] Therefore, how to build a railway engineering construction section-level platform that covers the entire chain of perception, transmission, processing, analysis, and decision-making, and can break down data barriers and achieve intelligent management and control, is a technical problem that urgently needs to be solved. Summary of the Invention

[0006] To address the shortcomings of existing technologies in railway construction section management, such as lagging data collection, passive risk control, and weak decision support, this invention provides a method, system, terminal equipment, and medium for constructing a railway engineering construction section-level platform based on time-series prediction and visual inspection networks. This invention constructs a five-layer intelligent platform and utilizes AI technology to achieve automatic data collection, cleaning, analysis, and decision-making, enabling refined and intelligent management and control of the entire railway engineering construction process.

[0007] In a first aspect, the present invention provides a method for constructing a railway engineering construction section-level platform based on a time-series prediction and visual inspection network, the method comprising: Based on the geographical grid division of the construction section, a multimodal sensor cluster is deployed to collect real-time data on on-site personnel behavior, construction equipment, construction environment parameters, and the real-time status of construction entity quality, thereby obtaining multi-source heterogeneous real-time status data to construct a global perception layer. Edge-side preprocessing is performed on the collected multi-source heterogeneous real-time status data. The preprocessed data is then spatiotemporally and semantically aligned with the BIM model to generate digital twin mapping data, thereby constructing a data processing layer. The edge-side preprocessing includes: protocol parsing, outlier cleaning, and keyframe extraction. The digital twin mapping data is deeply analyzed based on a pre-trained AI model library to output structured decision results in order to build an intelligent engine layer. The AI ​​model library includes: a progress prediction model based on LSTM time series network, a quality defect identification model based on YOLOv8 visual network, or a multi-factor safety risk early warning model based on random forest algorithm. Configure a business rules engine to transform the structured decision results into standardized business control instructions, and distribute the business control instructions to the corresponding field control equipment or business subsystems to achieve closed-loop control of the construction process, thereby building a business application layer; A role-based access control mechanism is established to create a correlation index for data at each level, providing differentiated visual interactive interfaces and functional permissions for different participants, thereby constructing a user interaction layer.

[0008] In one implementation, the preprocessed data is spatiotemporally and semantically aligned with the BIM model, including: Microsecond-level time synchronization of multi-source heterogeneous real-time status data is achieved using the Network Time Protocol. A BIM and GIS fusion engine is constructed to map multi-source heterogeneous real-time status data containing geographic coordinate information to the three-dimensional spatial coordinate system of the BIM model; Establish a mapping table between sensor IDs and BIM components, and attach multi-source heterogeneous real-time status data (non-spatial attributes) to the attributes of the corresponding BIM components to complete data fusion; Based on the construction schedule of railway engineering, the binding logic between the sensor ID and BIM components is dynamically adjusted according to the time nodes of the construction task book through a soft mapping mechanism, so that the same physical sensor is associated with the corresponding monitoring identity in different time windows.

[0009] In this way, by combining microsecond-level time synchronization, BIM+GIS fusion mapping and a soft mapping mechanism that evolves with the construction task cycle, the problem of mapping failure due to frequent changes in perception targets during long-cycle and multi-process construction of large-scale railway projects is solved. This achieves real-time and accurate anchoring of perception data and the identity of three-dimensional components, and constructs a digital twin that can automatically evolve with the physical progress.

[0010] In one implementation, the edge-side preprocessing includes: Outlier noise points in multi-source heterogeneous real-time state data are removed using a sliding window filtering algorithm. The video stream data in the multi-source heterogeneous real-time status data is filtered using an image sharpness evaluation algorithm. Video frames with sharpness below a preset threshold are discarded, key frames are extracted, and the extracted key frames are compressed. An adaptive sampling rate adjustment strategy based on data volatility is implemented. The dynamic intensity of the on-site physical characteristics is assessed by calculating the rate of change of the sensed values ​​in real time, and the sampling frequency is dynamically switched between a preset high-frequency range and a low-frequency range according to the rate of change.

[0011] In this preferred approach, traditional filtering and denoising, image screening, and adaptive sampling strategies based on data volatility are organically integrated. This enables intelligent allocation of sensing resources according to the dynamic intensity of the on-site physical characteristics. While not missing key features of instantaneous qualitative changes, it significantly reduces the transmission load of massive heterogeneous data at the section level, achieving a non-obvious balance between sensing accuracy and energy efficiency.

[0012] In one implementation, the operation of the multi-factor security risk early warning model based on the random forest algorithm includes: Define a risk assessment index system, which includes construction entity risk factors and construction equipment risk factors; The weight coefficients of construction entity risk factors and construction equipment risk factors were determined using the analytic hierarchy process (AHP). The normalized real-time monitoring values ​​of construction entity risk factors and construction equipment risk factors are input into the multi-factor safety risk early warning model for classification and reasoning, and the comprehensive risk level of the current construction area is output. Based on the comprehensive risk level and in accordance with the preset risk classification and response mechanism, a structured decision result containing the location of the risk source, the consequences of the risk, and disposal suggestions is automatically generated. Specifically, based on the comparison results of environmental parameters monitored by the global perception layer with preset sensitivity thresholds, the component defect results identified by the quality defect identification model, and / or the resource allocation suggestions generated by the progress prediction model, the input features and / or weight parameters in the multi-factor safety risk early warning model are corrected in real time.

[0013] By introducing real-time feedback correction logic of environment, quality and schedule parameters into the random forest early warning model, isolated safety monitoring is upgraded to multi-dimensional correlation risk assessment, which can sensitively capture the chain safety risks caused by the compression of the construction period or quality defects, and realize accurate quantitative prediction of potential systemic risks in complex construction environments.

[0014] In one implementation, the operation of the progress prediction model based on LSTM time series networks includes: Input historical construction log data, resource input data, and environmental weather data; The temporal dependency features of construction progress are extracted by using an LSTM temporal network to predict the amount of work to be completed within a preset time window in the future. The deviation between the predicted project completion amount and the planned project duration is calculated. If the deviation exceeds the threshold, a structured decision result is generated, including a schedule delay warning and resource allocation suggestions. The operation process also includes: introducing prior constraints based on the construction organization logic of railway engineering, and performing logical consistency verification on the predicted engineering completion quantity according to the logical relationship of the components in the BIM model, so as to eliminate prediction fluctuations that do not conform to the construction organization logic.

[0015] Based on the LSTM prediction model, prior constraints of railway construction organization logic and BIM component association verification are introduced to solve the technical defects of general AI models that are prone to non-physical logic fluctuations under interference data. This ensures that the prediction results strictly conform to the objective laws of railway engineering construction and greatly improves the technical reliability of schedule management under extreme working conditions.

[0016] In one implementation, business control commands are distributed to corresponding field control equipment or business subsystems to achieve closed-loop control of the construction process, including: A closed-loop feedback mechanism is constructed to track the execution status of the business control instructions and the results of on-site rectification in real time after the business control instructions are distributed to the corresponding field control equipment or business subsystems. The business application layer also has a built-in instruction arbitration mechanism based on priority weights, which is used to perform conflict assessment, collaborative optimization and execution reordering of the business control instructions according to a preset dynamic weight matrix of security, quality and schedule when there are logical conflicts in the instructions output by multiple AI models.

[0017] By establishing a priority-weighted instruction arbitration mechanism and a closed-loop feedback system, the problem of logical conflict in AI decision-making instructions under multi-objective management was solved. An automated execution boundary with safety and quality as the red line was established, ensuring that the construction site can always make the logically optimal and compliant autonomous response when facing multi-dimensional decision-making pressure.

[0018] In one implementation, the global perception layer further includes the following features: executing dynamic grid evolution logic based on construction risk distribution, extracting multi-source heterogeneous real-time state data within each grid and calculating grid risk entropy values, and when the risk entropy value exceeds a preset stability threshold, recursively subdividing the basic grid of the corresponding area into micro-grids, and adjusting the acquisition parameters of the multi-modal sensor cluster within the corresponding area in a coordinated manner.

[0019] By utilizing the dynamic grid evolution logic based on risk entropy, the spontaneous reorganization of sensing resources according to the distribution of on-site risks is realized, which solves the problem of uneven sensing coverage caused by the large number of points and long lines in railway sections. This enables the system to spontaneously form an ultra-fine monitoring field in high-risk areas, such as geologically fractured zones, and achieve high-magnification and accurate capture of the risk evolution process.

[0020] Secondly, embodiments of the present invention also provide a railway engineering construction section-level platform construction system based on temporal prediction and visual inspection networks, wherein the system is used to implement the steps of the railway engineering construction section-level platform construction method based on temporal prediction and visual inspection networks described in any of the above claims, and the system includes: The global perception layer construction module is used to deploy a multimodal sensor cluster based on the geographical grid division of the construction section, and to collect real-time data on the behavior of on-site personnel, construction equipment, construction environment parameters and the real-time status of construction entity quality, so as to obtain multi-source heterogeneous real-time status data. The data processing layer construction module is used to perform edge-side preprocessing on the collected multi-source heterogeneous real-time status data, and to perform spatiotemporal semantic alignment between the preprocessed data and the BIM model to generate digital twin mapping data. The edge-side preprocessing includes: protocol parsing, outlier cleaning and keyframe extraction. The intelligent engine layer construction module is used to perform deep analysis on the digital twin mapping data based on the pre-trained AI model library and output structured decision results. The AI ​​model library includes: a progress prediction model based on LSTM temporal network, a quality defect identification model based on YOLOv8 visual network, or a multi-factor safety risk early warning model based on random forest algorithm. The business application layer construction module is used to configure the business rule engine, transform the structured decision results into standardized business control instructions, and distribute the business control instructions to the corresponding field control equipment or business subsystems to achieve closed-loop control of the construction process. The user interaction layer module is used for role-based access control mechanisms, establishing correlation indexes for data at each level, and providing differentiated visual interactive interfaces and functional permissions for different participants.

[0021] Thirdly, embodiments of the present invention also provide a terminal device, wherein the terminal device includes a memory, a processor, and a railway engineering construction section-level platform construction program based on a time-series prediction and visual inspection network stored in the memory and executable on the processor. When the processor executes the railway engineering construction section-level platform construction program based on a time-series prediction and visual inspection network, it implements the steps of the railway engineering construction section-level platform construction method based on a time-series prediction and visual inspection network in any of the above-mentioned schemes.

[0022] Fourthly, embodiments of the present invention also provide a computer-readable storage medium, wherein the computer-readable storage medium stores a railway engineering construction section-level platform construction program based on a time-series prediction and visual inspection network, the railway engineering construction section-level platform construction program based on a time-series prediction and visual inspection network implementing the steps of the railway engineering construction section-level platform construction method based on a time-series prediction and visual inspection network as described in any of the above-mentioned schemes on the computer-readable storage medium.

[0023] Compared with existing technologies, this invention has the following significant advantages through deep empowerment of a five-layer architecture and scenario-based adaptation of AI technology: (1) End-to-End Data Closed Loop: Through a five-layer architecture design, a complete data closed loop is achieved from perception, transmission, processing, analysis to application, breaking down data silos in traditional construction management and ensuring data real-time performance and consistency. Based on this, the invention, through a closed-loop feedback mechanism and even automatic annotation logic, realizes the automatic conversion of rectification results into sample data and online incremental training of AI models, enabling the platform to self-evolve with evolving working conditions, ensuring continuous optimization of control logic and continuous improvement in decision-making accuracy.

[0024] (2) Intelligent Decision Support: By introducing various advanced AI models such as LSTM, YOLOv8, and Random Forest, in-depth analysis and prediction of multiple dimensions such as progress, quality, and safety are achieved, transforming traditional "experience-based decision-making" into "data-driven decision-making," significantly improving management efficiency and accuracy. Furthermore, prior constraints based on the construction organization logic of railway engineering are introduced, effectively eliminating prediction fluctuations caused by non-physical logic. At the same time, by establishing a three-in-one correlation evaluation matrix of quality, progress, and safety, parameter feedback and logical coupling between different AI models are realized, upgrading traditional isolated monitoring into intelligent auxiliary decision-making with in-depth industry knowledge and cross-dimensional collaboration.

[0025] (3) Refined Risk Management: Through a multi-factor fusion safety risk early warning model, the system achieves quantitative assessment and graded early warning of complex risk factors at the construction site, transforming passive response into proactive prevention and effectively reducing the incidence of construction safety accidents. At the same time, it innovatively proposes a dynamic grid evolution logic based on risk entropy, enabling the spontaneous focusing of sensing resources on high-risk areas; coupled with a command arbitration mechanism based on priority weights, it solves the problem of decision-making conflicts under multi-objective management, ensuring the timeliness, logic, and safety of risk handling under extreme and complex working conditions.

[0026] (4) Digital Twin Fusion: By deeply integrating real-time monitoring data with the BIM model through the data processing layer, a digital twin of the railway construction section was constructed, providing managers with an intuitive and visualized panoramic control perspective. A soft mapping mechanism that dynamically evolves with the construction process solved the mapping problem of frequently changing perception targets during long-cycle construction. Combined with knowledge graph-based semantic retrieval and virtual peer-to-peer collaboration technology, it provided managers with a panoramic control perspective that transcends traditional visual displays and possesses deep causal logic, greatly improving the efficiency of collaborative management with multiple parties involved. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating a preferred embodiment of the railway engineering construction section-level platform construction method based on time-series prediction and visual inspection networks provided by the present invention.

[0028] Figure 2 This is a flowchart illustrating the spatiotemporal semantic alignment process in the railway engineering construction section-level platform construction method based on temporal prediction and visual inspection networks provided in this embodiment of the invention.

[0029] Figure 3 This is a schematic diagram illustrating the operation flow of the progress prediction model based on LSTM time series network in the railway engineering construction section-level platform construction method based on time series prediction and visual inspection network provided in the embodiments of the present invention.

[0030] Figure 4 This is a schematic diagram illustrating the operation of the multi-factor safety risk early warning model based on the random forest algorithm in the railway engineering construction section-level platform construction method based on time-series prediction and visual inspection network provided in the embodiments of the present invention.

[0031] Figure 5 This is an architecture diagram of a railway engineering construction section-level platform construction system based on temporal prediction and visual inspection networks, provided in an embodiment of the present invention.

[0032] Figure 6 A schematic diagram of a terminal device provided in an embodiment of the present invention. Detailed Implementation

[0033] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0034] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content, operations, or steps, nor does it require execution in the described order. For example, some operations or steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0035] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0036] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present invention, the terms "first" and "second" are used in the embodiments of the present invention to distinguish identical or similar items with essentially the same function and effect. For example, "first control information" and "second control information" are only used to distinguish different control information and do not limit their order.

[0037] Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or the order of execution, and that the words "first" and "second" do not necessarily imply that they are different.

[0038] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0039] To address the problems of existing technologies, this embodiment provides a method for constructing a railway engineering construction section-level platform based on temporal prediction and visual inspection networks. Based on this method, a five-layer architecture design achieves a closed-loop data flow from perception, transmission, processing, analysis to application, breaking down data silos in traditional construction management and ensuring data real-time performance and consistency. In specific applications, this embodiment first deploys a multimodal sensor cluster based on the geographical grid division of the construction section. This cluster collects real-time data on on-site personnel behavior, construction equipment, construction environment parameters, and the real-time status of construction entities, obtaining multi-source heterogeneous real-time status data to construct a comprehensive perception layer. Then, edge-side preprocessing is performed on the collected multi-source heterogeneous real-time status data. The preprocessed data is then spatiotemporally and semantically aligned with the BIM model to generate digital twin mapping data, constructing a data processing layer. The edge-side preprocessing includes protocol parsing, outlier cleaning, and keyframe extraction. Next, a deep analysis of the digital twin mapping data is performed based on a pre-trained AI model library to output structured decision results, thus constructing an intelligent engine layer. This AI model library includes: a progress prediction model based on an LSTM temporal network, a quality defect identification model based on a YOLOv8 visual network, or a multi-factor safety risk early warning model based on a random forest algorithm. Then, a business rule engine is configured to transform the structured decision results into standardized business control instructions, which are then distributed to corresponding field control equipment or business subsystems to achieve closed-loop control of the construction process, thus constructing a business application layer. Finally, a role-based access control mechanism is implemented to establish a correlation index for data at each level, providing differentiated visual interactive interfaces and functional permissions for different participants, thus constructing a user interaction layer.

[0040] The railway engineering construction section-level platform construction method based on temporal prediction and visual inspection networks in this embodiment can be applied to terminal devices, which can be intelligent products such as computers. Figure 1 As shown in the figure, the construction method of the railway engineering construction section-level platform based on time-series prediction and visual inspection network in this embodiment includes the following steps: Step S100: Based on the geographical grid division of the construction section, deploy a multimodal sensor cluster to collect real-time data on on-site personnel behavior, construction equipment, construction environment parameters, and the real-time status of construction entity quality, thereby obtaining multi-source heterogeneous real-time status data to construct a global perception layer.

[0041] This step is used to construct a global perception layer to collect multi-source heterogeneous real-time status data of the construction site. The global perception layer is used to deploy a multimodal sensor cluster based on the geographical grid division of the construction section, collecting real-time data on personnel behavior, construction equipment, construction environment parameters, and the quality of construction entities, thus obtaining multi-source heterogeneous real-time status data. This step aims to solve the problems of incomplete, inaccurate, and untimely data collection in traditional construction management. Specifically, to achieve refined perception coverage of the construction section, this embodiment combines factors such as the topography of the construction section, the type of construction area (e.g., roadbed section, bridge section, tunnel section, station section, etc.), and construction progress to divide the geographical grid. In practical application, this embodiment uses a 10m × 10m basic grid. For critical structural areas such as bridges and tunnels, a 5m × 5m grid can be used for denser division; for auxiliary areas such as construction access roads and material storage yards, a simplified 20m × 20m grid can be used. Next, grid coding is performed. This embodiment uses a 16-bit coding rule of "section number - area type code - grid sequence number". For example, "BJ-QL-000123" represents grid number 123 of the bridge area (QL) in a certain city section (BJ). When the scope of the construction area changes (such as adding a new construction surface or a completed and handed-over area) or the construction focus shifts, the grid division can be automatically updated by the platform or manually configured and adjusted to ensure the effectiveness of the sensing coverage.

[0042] The multimodal sensor cluster in this embodiment includes four main categories of sensors: personnel sensing, equipment monitoring, environmental monitoring, and quality inspection. Based on the grid function and construction requirements, it can be deployed on demand and achieve full coverage. The specific deployment scheme is shown in Table 1.

[0043] Table 1 In a preferred embodiment, the global perception layer also has the ability to dynamically adjust the grid division fineness based on the distribution of construction risks. This embodiment extracts multi-source heterogeneous real-time state data within each grid and uses information entropy theory to calculate the risk entropy value of that grid under a specific process. When the risk entropy value within a grid exceeds a preset stability threshold, such as when tunnel excavation enters a fault fracture zone or when a bridge undergoes large-segment hoisting, the global perception layer automatically initiates grid evolution logic, recursively subdividing the basic grid of that area, for example, refining it from 10m × 10m to a 2m × 2m microgrid. Simultaneously with grid refinement, the acquisition parameters of the multimodal sensor cluster within that area are adjusted to ensure higher spatial and temporal accuracy in capturing high-risk evolution processes, thereby achieving precise allocation of sensing resources to high-risk areas.

[0044] To further improve the response speed and energy efficiency of this invention under complex working conditions, an event-driven cross-modal collaborative wake-up mechanism is integrated within the global perception layer. Under this mechanism, environmental monitoring sensors, such as temperature, humidity, and rainfall sensors, are normally in a low-power polling state. However, when equipment monitoring sensors detect abnormal mechanical vibrations through edge-side analysis, or personnel sensing sensors identify unauthorized entry into a restricted area grid, this embodiment immediately generates a wake-up interrupt command. This command instantly activates the high-definition industrial cameras and ultrasonic flaw detectors within the grid and adjacent grids via the local sensing network, performing high-frame-rate image capture or internal defect scanning. This intermodal logical triggering relationship transforms full-time, full-volume acquisition into target-oriented acquisition, significantly reducing the communication load and data processing pressure of large-scale sensor clusters at the section level while ensuring no critical data is missed.

[0045] For special scenarios in railway engineering, such as tunnels and deep foundation pits where Global Navigation Satellite System (GNSS) signals are limited or blocked, the global perception layer adopts a relative collaborative positioning strategy based on construction equipment anchor points. Within a closed grid where satellite signal strength is below a preset threshold, this embodiment can automatically switch to a relative coordinate reconstruction mode. In this mode, the global perception layer selects heavy construction equipment with fixed three-dimensional spatial coordinates, already aligned in the Building Information Model (BIM), such as tunnel boring machines and intelligent grouting pump trucks, as dynamic reference anchor points. By using ultra-wideband (UWB) communication modules integrated into personnel smart wristbands and safety helmets to perform real-time ranging with the anchor points, the relative spatial coordinates of personnel and small machinery can be reconstructed within a local grid using trilateration algorithms and mapped back to the global geographic coordinate system, ensuring the positioning continuity of the perception network in extreme environments such as long and deep tunnels.

[0046] Furthermore, to ensure the data uploaded to the intelligent engine layer possesses high-quality analytical value, the global perception layer implements an adaptive sampling rate adjustment strategy based on data volatility. Each sensor node incorporates primary data evaluation logic, assessing the dynamic intensity of on-site physical characteristics by calculating the first derivative or variance of the sensed values ​​in real time. Taking a concrete strength sensor as an example, during the initial setting and hardening stage after pouring, when the strength growth rate is rapid, the sensor automatically increases the sampling frequency to a preset high-frequency range; while when the concrete enters a period of stable strength growth or the later stages of curing, the sampling frequency decreases to a low-frequency range. This adaptive adjustment ensures that sufficiently rich feature points can be recorded at critical moments of qualitative change in technical parameters, while automatic noise reduction and compression of data are achieved during the stable phase. Preferably, the aforementioned sensors can also incorporate a local cache module, allowing data to be stored locally when the network is interrupted and automatically retransmitted after network recovery, preventing data loss.

[0047] The above deployment scheme is used to deploy a multimodal sensor cluster to collect real-time data on on-site personnel behavior, construction equipment, construction environment parameters, and the quality of construction entities in a real-time and energy-efficient manner, thereby obtaining multi-source heterogeneous real-time status data.

[0048] Step S200: Perform edge-side preprocessing on the collected multi-source heterogeneous real-time status data, and align the preprocessed data with the BIM model in a spatiotemporal semantic manner to generate digital twin mapping data to construct a data processing layer. The edge-side preprocessing includes: protocol parsing, outlier cleaning, and keyframe extraction.

[0049] This step is used to construct the data processing layer, enabling edge-side preprocessing of multi-source heterogeneous real-time status data. Because the data collected by the perception layer is multi-source, heterogeneous, massive, and high-noise, it must be processed before it can be used by AI models. Specifically, the edge-side preprocessing in this embodiment includes: protocol parsing, outlier cleaning, and keyframe extraction. This embodiment incorporates a multi-protocol parsing engine in the edge computing gateway, supporting unified parsing of transmission protocols from different sensors, enabling unified networking of heterogeneous devices. This embodiment can define standardized parsing templates for each transmission protocol, and then uniformly convert the parsed multi-source heterogeneous real-time status data into JSON format (a lightweight data exchange format). Fields include sensor ID, acquisition timestamp, grid code, data type, data value, and data status. This embodiment also supports dynamic protocol expansion; when a new sensor adopts a new protocol, a new protocol parsing template can be uploaded through the platform without modifying the edge gateway hardware configuration.

[0050] Furthermore, this embodiment can utilize a sliding window filtering algorithm to remove outlier noise points from multi-source heterogeneous real-time status data, achieving outlier cleanup. For example, the sliding window filtering algorithm can be used to remove spike outliers caused by electromagnetic interference from sensors. The sliding window filtering algorithm can set the window size to 5 (i.e., 5 consecutive data points constitute one window), then use the average value of the data within the window as the output value, and then compare the average value of the output of each window to identify and remove outlier noise points, thereby achieving clear outlier removal.

[0051] In a preferred embodiment, the data processing layer integrates an outlier detection algorithm based on construction conditions during edge-side preprocessing. Addressing the non-stationary noise generated by strong electromagnetic interference and severe mechanical vibration at railway construction sites, the edge computing gateway not only utilizes sliding window filtering for basic noise reduction but also dynamically adjusts the width and weighting factors of the filtering window based on the current operating status of the construction machinery, such as startup, full load, and idling. When the signal frequency captured by the vibration sensor highly overlaps with the operating frequency envelope of the construction machinery, this embodiment identifies it as valid operational data rather than noise points, thus avoiding the erroneous deletion of high-frequency valid features by traditional filtering algorithms.

[0052] In addition, for the outliers that have been removed, this embodiment can also use linear interpolation to complete them. That is, based on two valid data points before and after the outlier, the complete value is calculated by fitting, so as to ensure the continuity of the data.

[0053] Furthermore, this embodiment utilizes an image sharpness evaluation algorithm to filter video stream data from multi-source heterogeneous real-time status data, discarding video frames with sharpness below a preset threshold, extracting keyframes, and compressing the extracted keyframes to ensure reduced storage usage while maintaining image quality. Simultaneously, the keyframe extraction logic incorporates a filtering mechanism based on temporal saliency detection. When unexpected visually significant events such as sudden changes in material location or unauthorized personnel entry occur in the video stream, this embodiment automatically increases the sampling resolution and extends the coverage duration of the keyframe sequence, ensuring the feature completeness of subsequent AI model analysis.

[0054] Furthermore, this embodiment includes the following steps when performing spatiotemporal semantic alignment: Step S201: Use the network time protocol to perform microsecond-level time synchronization of multi-source heterogeneous real-time status data; Step S202: Construct a BIM and GIS fusion engine to map multi-source heterogeneous real-time status data containing geographic coordinate information to the three-dimensional spatial coordinate system of the BIM model; Step S203: Establish a mapping table between sensor IDs and BIM components, and mount the multi-source heterogeneous real-time status data (non-spatial attributes) to the attributes of the corresponding BIM components to complete data fusion.

[0055] Specifically, this embodiment performs spatiotemporal semantic alignment between the preprocessed data and the BIM model (Building Information Modeling) to generate digital twin mapping data. Spatiotemporal semantic alignment is the core of achieving accurate mapping between the physical world and the digital twin model. Through three main steps—time synchronization, spatial coordinate mapping, and attribute mounting—deep integration of multi-source data and the BIM model is achieved. Specifically, as... Figure 2As shown in the diagram, this embodiment utilizes the Network Time Protocol (NTP) to perform microsecond-level time synchronization of multi-source heterogeneous real-time status data, ensuring consistent timestamps across all data. Next, a BIM and GIS (Geographic Information System) fusion engine is constructed to map the multi-source heterogeneous real-time status data containing geographic coordinate information to the three-dimensional spatial coordinate system of the BIM model. In the specific mapping process, this embodiment uses topographic information such as terrain, roads, and water systems from the GIS data as the background layer of the BIM model, and overlays data such as the grid division results of the construction area and sensor deployment locations as vector layers onto the BIM model, achieving an integrated presentation of geographic and engineering information. Finally, this embodiment establishes a mapping relationship table between sensor IDs and BIM components, attaching non-spatial attribute multi-source heterogeneous real-time status data to the attributes of the corresponding BIM components, completing data fusion, and generating digital twin mapping data. The structure of the mapping relationship table in this embodiment includes fields such as sensor ID, BIM component ID, component name, component type, data type, attached attribute fields, and update frequency, as shown in Table 2.

[0056] Table 2 Once the sensors collect new data and complete preprocessing, the corresponding attribute values ​​of the BIM components are updated in real time according to the mounting attribute fields in the mapping table, thereby achieving dynamic linkage between data and the model.

[0057] Furthermore, to achieve a deep and dynamic depiction of the physical construction process, the mapping table in this embodiment can automatically evolve according to the changes in the construction progress. Considering the distinct phased characteristics of railway engineering, such as the transition from the early stage of roadbed and bridge construction to the later stage of electrical, electronic, and telecommunications installation, the correspondence between sensor IDs and BIM components is not rigidly fixed. Instead, it is logically switched according to the time nodes in the construction task book through a built-in soft mapping mechanism. This mechanism allows the same physical sensor to be assigned different monitoring identities in different time windows, thereby ensuring that the sensing data can always be accurately attached to the most relevant work components or process nodes at the moment.

[0058] Based on updating the real-time attributes of BIM components, this embodiment also utilizes a multi-dimensional vector space to record the evolution trajectory of these physical parameters and constructs a logical topological map covering components, environment, and processes in the digital space by calculating their spatiotemporal correlation. This allows for the identification of causal relationships across components from isolated data points, such as the impact of changes in environmental humidity on the strength growth rate of a specific component. This provides deeper logical support for subsequent intelligent analysis, achieving a technological leap from simple location mapping to deep semantic fusion. Step S300: Based on a pre-trained AI model library, perform deep analysis on the digital twin mapping data and output structured decision results to construct an intelligent engine layer. The AI ​​model library includes: a progress prediction model based on an LSTM temporal network, a quality defect identification model based on a YOLOv8 visual network, or a multi-factor safety risk early warning model based on a random forest algorithm.

[0059] This step is used to build the intelligent engine layer, which is the brain of the platform and integrates various AI models optimized for railway construction scenarios. In this embodiment, the AI ​​model library can be deployed collaboratively on the cloud and edge. For example, the progress prediction model and the multi-factor safety risk early warning model are deployed on the cloud to support batch data processing and complex calculations, while the quality defect identification model is deployed on the edge to support real-time processing of video stream data.

[0060] In this embodiment, the progress prediction model based on an LSTM (Long Short-Term Memory) temporal network is used to predict the amount of work to be completed within a preset time window based on construction data and real-time status data, identify progress deviations, and generate resource allocation suggestions. The LSTM temporal network in this embodiment adopts a structure of input layer, hidden layer, dropout layer (layer used to reduce overfitting), and output layer.

[0061] The operation process of the schedule prediction model in this embodiment includes the following steps: Step S301: Input historical construction log data, resource input data, and environmental weather data; Step S302: Extract the temporal dependency features of the construction progress through an LSTM temporal network to predict the amount of work to be completed within a future preset time window; Step S303: Calculate the deviation between the predicted project completion amount and the planned project duration. If the deviation exceeds the threshold, generate a structured decision result that includes a schedule delay warning and resource allocation suggestions.

[0062] like Figure 3As shown, when predicting the progress, the input data includes historical construction log data, resource input data, and environmental weather data, all of which can be obtained from digital twin mapping data. The construction log data includes daily project completion volume (e.g., concrete pouring volume, number of rebar ties, tunnel excavation meters) and construction process completion status (e.g., start and end times of formwork, pouring, and curing). Resource input data includes daily number of construction workers, construction machinery hours, and material consumption (e.g., the amount of rebar, cement, and aggregate used). Environmental weather data includes daily average temperature, rainfall, wind speed, and sunshine duration. This embodiment can extract the temporal dependency features of the construction progress from the above input data through the hidden layers of an LSTM temporal network to predict the project completion volume within a preset time window. Then, the deviation between the predicted project completion volume and the planned construction period is calculated. If the deviation exceeds a threshold, a structured decision result is generated, including a progress delay warning and resource allocation suggestions, such as increasing the number of construction workers, adjusting equipment hours, and optimizing the material supply plan.

[0063] This embodiment uses a YOLOv8 visual network (an algorithm framework supporting object detection and image segmentation) to automatically identify the type, location, and size of quality defects by acquiring surface images and video stream data of construction entities from high-definition cameras, and then generates quality rectification instructions. This embodiment defines quality defects as follows: concrete component defects (such as cracks, honeycombing, pitting, exposed reinforcement, and holes), steel structure defects (such as weld cracks, corrosion, and deformation), and tunnel lining defects. When constructing the dataset, quality defect images and video data from multiple railway construction sections were collected, and annotation tools were used to mark defect areas with rectangular boxes, including defect type, boundary coordinates, and defect size. Based on the constructed dataset, the YOLOv8 visual network was transferred and fine-tuned to obtain a quality defect identification model adapted to the railway engineering quality defect identification scenario. This quality defect identification model can analyze on-site monitoring videos or drone images in real time from digital twin mapping data, automatically select defect areas, identify defect types, calculate defect confidence, and output structured decision results. At this time, the structured decision results include information such as defect ID, defect type, confidence, defect location, defect size, and identification time, and push the structured decision results to the business application layer.

[0064] In this embodiment, a multi-factor risk assessment model based on random forests is used to achieve comprehensive risk level evaluation and early warning for the construction area based on multi-dimensional risk factors of construction entities and construction equipment, and to generate risk management suggestions. This embodiment first defines a risk assessment index system, which includes construction entity risk factors and construction equipment risk factors. Construction entity risk factors may include concrete strength, support settlement, etc., while construction equipment risk factors may include equipment operating temperature, equipment failure frequency, etc. Then, the weight coefficients of the construction entity risk factors and construction equipment risk factors are determined using the analytic hierarchy process (AHP). Specifically, this embodiment determines the corresponding weight coefficients based on the importance of the construction entity risk factors and construction equipment risk factors. This embodiment preferably uses the weight coefficients of the construction entity risk factors. The weighting coefficient of the risk factor greater than that of construction equipment During model training, this embodiment can collect historical safety risk data from multiple railway construction sections. Each sample in this historical safety risk data includes historical measurements of construction entity risk factors and construction equipment risk factors, along with their corresponding actual risk levels, thus forming a dataset. Based on this dataset, the random forest algorithm is configured and trained to obtain a multi-factor risk assessment model adapted to this invention.

[0065] The operation process of the multi-factor security risk early warning model in this embodiment includes the following steps: Step S31: Define a risk assessment index system, which includes construction entity risk factors and construction equipment risk factors; Step S32: Use the analytic hierarchy process (AHP) to determine the weight coefficients of the construction entity risk factor and the construction equipment risk factor, wherein the weight coefficient of the construction entity risk factor is greater than the weight coefficient of the construction equipment risk factor. Step S33: Input the real-time monitoring values ​​of the normalized construction entity risk factors and construction equipment risk factors into the multi-factor safety risk early warning model for classification and reasoning, and output the comprehensive risk level of the current construction area. Step S34: Based on the comprehensive risk level and in accordance with the preset risk classification response mechanism, automatically generate a structured decision result that includes the location of the risk source, the consequences of the risk, and disposal suggestions.

[0066] In practical applications, such as Figure 4As shown in the diagram, this embodiment first defines a risk assessment index system. Then, the weight coefficients of the construction entity risk factors and construction equipment risk factors are determined using the analytic hierarchy process (AHP). Next, real-time monitoring values ​​of the construction entity risk factors and construction equipment risk factors are obtained from digital twin mapping data. These normalized real-time monitoring values ​​are then input into a multi-factor safety risk early warning model for classification and reasoning to output the comprehensive risk level of the current construction area. In practical applications, this embodiment can also directly output risk values ​​through the multi-factor safety risk early warning model and then determine the corresponding comprehensive risk level based on these values. Finally, based on the comprehensive risk level and according to a preset risk grading response mechanism, a structured decision result containing the location of the risk source, risk consequences, and disposal recommendations is automatically generated.

[0067] Furthermore, in a preferred embodiment, the AI ​​model library within the intelligent engine layer is not characterized by independent operation of individual models, but rather by multi-dimensional logical coupling achieved through data sharing and weight feedback mechanisms. This embodiment establishes a three-in-one correlation evaluation matrix based on quality, schedule, and safety, enabling the output results of each model to correct the input features or weight parameters of other models in real time. For example, when a quality defect identification model based on the YOLOv8 visual network identifies severe honeycombing or cracks in critical components, such as tunnel initial supports or bridge bearings, the identification result is input as a feature variable in real time into the multi-factor safety risk early warning model, automatically triggering dynamic weighted calculation of risk factors for the construction entity, thereby increasing the comprehensive risk level of the grid area. Similarly, when a schedule prediction model based on the LSTM temporal network identifies a serious delay in the current schedule and generates a resource allocation suggestion to increase equipment investment, this decision information is synchronously fed back to the safety risk early warning model, guiding the system to automatically increase the monitoring weight of risk factors for construction equipment to prevent equipment fatigue or overload risks caused by rushing the schedule. This deep collaboration mechanism between models enables the platform to make conflict identification and optimal decisions from a global perspective.

[0068] To address the extreme working conditions commonly encountered in railway construction, specific algorithm optimizations were performed on each model in this embodiment. For the quality defect identification model, considering the low-light and high-dust imaging environment inside tunnels, an adaptive image enhancement module was integrated into the YOLOv8 network architecture. By performing contrast stretching and denoising preprocessing on the original video stream, the accuracy of identifying minute cracks in concrete lining under low-light conditions was significantly improved. For the progress prediction model, the LSTM network introduced prior constraints based on the construction organization logic of railway engineering when processing temporal features. The system performs logical consistency checks on the predicted project completion quantities based on the logical relationships between components in the BIM model, such as the construction order of the lower structure followed by the upper structure, eliminating prediction fluctuations that do not conform to construction specifications. This makes the progress prediction results more closely reflect the operational characteristics of railway engineering, which involves numerous points, long lines, and tight procedures.

[0069] Furthermore, during the operation of the multi-factor safety risk early warning model based on the random forest algorithm, the weight coefficients determined by the analytic hierarchy process are not static but have adaptive adjustment capabilities based on real-time operating conditions. This embodiment incorporates an environmental impact factor mapping table. When environmental parameters monitored by the full-domain perception layer, such as wind speed, rainfall, and gas concentration, exceed preset sensitivity thresholds, the algorithm automatically increases the real-time weights of construction environmental factors and correspondingly reduces the weight ratios of construction equipment or personnel behavior to ensure that the early warning results can sensitively capture the systemic risks brought about by sudden environmental disasters. When outputting structured decision results, the intelligent engine layer can also use the business rule engine to collaboratively optimize the allocation suggestions of different models. For example, when facing a conflict between catching up on schedule and quality improvement decisions, it can automatically generate a phased execution sequence based on the principle of risk minimization, prioritizing ensuring quality standards are met before replenishing resources to catch up on schedule. This automated balancing based on algorithmic logic effectively avoids blind command caused by information asymmetry in traditional management models. Step S400: Configure the business rule engine to convert the structured decision results into standardized business control instructions, and distribute the business control instructions to the corresponding field control equipment or business subsystems to realize closed-loop control of the construction process, so as to build a business application layer.

[0070] This step is used to construct the business application layer, which configures the business rule engine to transform the structured decision results into standardized business control instructions and distribute these instructions to the corresponding on-site control equipment or business subsystems, achieving closed-loop control of the construction process. This embodiment first configures the business rule engine, which presets various business rules. For example, when the overall risk level is red, the power supply to the relevant area is immediately cut off and an evacuation broadcast is initiated; when a rebar spacing defect is identified, a quality rectification notice is automatically generated and pushed to the relevant responsible person's mobile APP. Based on this business rule engine, this embodiment can transform the structured decision results obtained in the above steps into standardized business control instructions and distribute these instructions to the corresponding on-site control equipment or business subsystems, achieving closed-loop control of the construction process. In practical applications, the business control instructions in this embodiment can be divided into quality rectification instructions, schedule adjustment instructions, safety handling instructions, and equipment control instructions. When distributing business control commands to corresponding field control devices or business subsystems, this embodiment can sample diverse distribution methods based on the command type and the recipient. For example, for field control devices (such as ventilation equipment and construction machinery), the business control command can be directly issued by calling the device's control interface through industrial Ethernet or 5G network to achieve automatic control of the device. For field personnel such as construction workers, quality inspectors, and safety officers, the business control command can be pushed through the platform's accompanying mobile APP.

[0071] In a preferred embodiment, the business application layer incorporates a priority-weighted instruction arbitration mechanism to resolve logical conflicts in multi-dimensional decision-making results at the execution end. Due to the extremely complex railway construction environment, instructions output by the intelligent engine layer may exhibit temporal or logical exclusivity. For example, a progress prediction model might suggest increasing equipment power to catch up with the schedule, while a safety risk warning model might suggest limiting equipment speed due to detected abnormal environmental parameters. In this case, the business rule engine does not directly forward the instructions but instead enters a conflict assessment phase. This embodiment can collaboratively optimize and reorder multiple control instructions based on a preset dynamic weight matrix, such as safety first, quality priority, and consideration of progress. If an instruction conflict involves life safety or significant quality hazards, the engine will automatically block lower-priority progress adjustment instructions and output a composite instruction containing conflict explanations and execution corrections to the distribution terminal, ensuring that the actions of on-site control equipment and business subsystems always remain within a logically safe range.

[0072] Furthermore, to address the potential instruction delays caused by long-distance transmission in railway sections, the business application layer supports predictive instruction issuance and edge autonomous closed-loop processing. For equipment control instructions with high timeliness requirements, the business rule engine predictively issues a pre-set rule set to the edge computing gateway before the risk value approaches a critical point, based on the parameter evolution trends in the digital twin mapping data. This allows the edge side to directly complete a small closed loop of analysis-control-feedback locally according to the pre-set rules when network fluctuations or momentary interruptions occur. After the network recovers, the entire process record data is synchronized back to the business application layer. This architecture, combining centralized decision-making and edge autonomy, significantly improves the legal stability and business continuity of the railway construction platform under extreme communication conditions. Further, this embodiment also constructs a closed-loop feedback mechanism. After distributing business control instructions to the corresponding field control equipment or business subsystems, the execution status of the business control instructions and the on-site rectification results are tracked in real time. The rectification result data is then fed back and labeled as sample data. This sample data is used for online incremental training and iterative optimization of the AI ​​model library, improving the analytical accuracy and decision-making rationality of the AI ​​model. In this process, preferably, machine vision-based automatic verification and automatic labeling logic is also integrated. When business control instructions are distributed and executed, such as rectification instructions for defects in rebar spacing, this embodiment will link and call the visual inspection network in the intelligent engine layer to perform a second automatic scan of the rectified grid area. If the visual inspection confirms that the defect has been eliminated, this embodiment will automatically aggregate the original monitoring data, intermediate decision parameters, and final rectification results in the closed-loop process, and use semantic analysis technology to label them with positive labels. If the rectification does not meet the standards, it will be labeled as a negative sample or a difficult sample. Before storing the returned data in the AI ​​model library for incremental training, the business application layer can also use a data filtering mechanism based on confidence thresholds to remove redundant or low-value samples, ensuring that only high-quality data that can characterize the specificity of the working condition or edge cases can participate in the iterative optimization of the model. Step S500: A role-based access control mechanism is established to create an association index for data at each level, providing differentiated visual interactive interfaces and functional permissions for different participants to build a user interaction layer.

[0073] This step is used to build the user interaction layer, which is used for role-based access control. For example, in this embodiment, the construction unit has full-function permissions, focusing on on-site work scheduling, hidden danger investigation and rectification, and resource inspection. The supervision unit has permissions for on-site supervision, process acceptance, and instruction issuance, focusing on viewing quality inspection data and rectification responses. The construction unit has macro-level supervision permissions, focusing on viewing overall progress, investment completion, and control of major risk sources. In addition, this embodiment also establishes a correlation index for data at each level to achieve rapid data query and traceability. Specifically, this includes: establishing a correlation index for sensor data, defect data, risk data, and BIM components based on the three-dimensional spatial coordinates of the BIM model, supporting quick querying of all relevant data for a component by clicking on it. Next, based on a unified timestamp, a time-related index is established for different types of data, supporting querying all construction data within a certain time period by time range, such as progress data, quality defect data, and safety warning data for a specific day. Finally, a business-related index is established for instruction data, execution result data, and sample data, supporting querying the corresponding execution result and sample data by instruction ID, achieving full traceability of the business process.

[0074] This embodiment can also provide differentiated visual interactive interfaces for different participants. It can employ an integrated interactive interface design combining 3D visualization, 2D statistics, and early warning alerts. For example, it can construct a large-screen command center for the entire project section based on GIS+BIM, displaying progress red flag maps, quality heat maps, safety risk cloud maps, and personnel and equipment trajectory maps in a 3D visualization manner to assist management in making macro-level decisions. Furthermore, this embodiment can integrate the platform with the project's existing labor real-name system, material weighbridge system, and mixing plant management system to achieve cross-system data linkage. For instance, when the progress prediction model indicates a delay, the system automatically calculates the required additional labor and raw material quantities and sends resource allocation suggestions to the corresponding systems.

[0075] In a preferred embodiment, the user interaction layer also features a dynamic adaptive adjustment function for the interface based on the safety and quality status of the construction site. This embodiment not only assigns permissions based on preset roles but also switches the display priority of the interactive interface in real time according to the risk level output by the intelligent engine layer. For example, when the comprehensive risk level of a certain grid area triggers a red alert, this embodiment automatically pushes an emergency command mode to the interfaces of all participating parties associated with that area, forcing the rendering of a digital twin real-time window of the risk area on the top layer of the GIS+BIM large screen and mobile APP, and temporarily locking irrelevant business operations until the safety handling instruction completes the closed-loop feedback. This event-priority-based interactive control logic ensures the absolute prominence of key information in massive project data, significantly improving the efficiency of emergency response under multi-party collaboration.

[0076] Furthermore, to address the challenge of accurately tracing massive amounts of fragmented data in railway engineering, a semantic retrieval and association display technology based on knowledge graphs is introduced in the user interaction layer. Building upon existing spatial, temporal, and business-related indexes, this embodiment utilizes a logical topology graph generated by the data processing layer to support cross-domain, penetrating queries by stakeholders using natural language or multi-dimensional composite conditions. For example, when a supervision unit clicks on a specific bridge pier BIM component, the interface not only displays the real-time strength data of the load but also automatically associates and displays the external environmental temperature and humidity curves during the component's construction, the vibration characteristic graphs of the construction machinery, and the relevant real-name information of the labor teams. This data clustering display based on causal logic transforms isolated monitoring indicators into technical records with an engineering context, providing in-depth digital evidence support for quality traceability and accident analysis.

[0077] Furthermore, the user interaction layer integrates a digital twin virtual peer-to-peer collaborative working mechanism for multi-party collaboration. Considering the large span and dispersed personnel of railway sections, this embodiment supports construction, supervision, and development units accessing the same 3D digital twin scene remotely via their respective terminals. Under this mechanism, interactive actions of different participants, such as red-circle markings, defect markers, or viewpoint panning on the BIM model, can be synchronized in real time and automatically recorded as collaborative evidence in the business association index. When the progress prediction model indicates a delay requiring resource allocation, this embodiment automatically simulates construction organization animations under different resource allocation schemes in the collaborative interface, assisting multiple parties in reaching optimal decisions through visual deduction. This functional leap from information display to collaborative decision-making greatly enhances practicality and legal stability in complex management scenarios.

[0078] In summary, this invention, by constructing a five-layer architecture consisting of a global perception layer, a data processing layer, an intelligent engine layer, a business application layer, and a user interaction layer, leverages AI technology to deeply empower railway engineering construction management. This addresses the persistent problems of data lag, risk control failure, and blind decision-making in traditional management models, providing strong technical support for the digital transformation of railway engineering construction.

[0079] Based on the above embodiments, the present invention also provides a railway engineering construction section-level platform construction system based on temporal prediction and visual inspection networks, the system being used to implement the steps in the above method embodiments. Figure 5As shown in the figure, the railway engineering construction section-level platform based on time-series prediction and visual inspection network in this embodiment includes: a global perception layer construction module 10, a data processing layer construction module 20, an intelligent engine layer construction module 30, a business application layer construction module 40, and a user interaction layer construction module 50. Specifically, the global perception layer construction module 10 is used to deploy a multimodal sensor cluster based on the geographical grid division of the construction section, and collect real-time data on the behavior of on-site personnel, construction equipment, construction environment parameters, and the real-time status of construction entity quality to obtain multi-source heterogeneous real-time status data. The data processing layer construction module 20 is used to perform edge-side preprocessing on the collected multi-source heterogeneous real-time status data, and perform spatiotemporal semantic alignment between the preprocessed data and the BIM model to generate digital twin mapping data. The edge-side preprocessing includes: protocol parsing, outlier cleaning, and keyframe extraction. The intelligent engine layer construction module 30 is used to perform deep analysis on the digital twin mapping data based on a pre-trained AI model library, and output structured decision results. The AI ​​model library includes: a progress prediction model based on an LSTM time series network, a quality defect identification model based on a YOLOv8 visual network, or a multi-factor safety risk early warning model based on a random forest algorithm. The business application layer construction module 40 is used to configure a business rule engine, convert the structured decision results into standardized business control instructions, and distribute these instructions to corresponding field control equipment or business subsystems to achieve closed-loop control of the construction process. The user interaction layer construction module 50 is used for a role-based access control mechanism, establishing a correlation index for data at each level, and providing differentiated visual interactive interfaces and functional permissions for different participants.

[0080] The railway engineering construction section-level platform construction system based on temporal prediction and visual inspection network in this embodiment is based on the same principle as the steps in the above method embodiments, and will not be repeated here.

[0081] Based on the above embodiments, the present invention also provides a terminal device, the principle block diagram of which can be as follows: Figure 6As shown, the terminal device includes a processor, memory, network interface, display screen, and temperature sensor connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminal devices via a network connection. When the computer program is executed by the processor, it implements a method for constructing a railway engineering construction section-level platform based on a time-series prediction and visual inspection network. The display screen can be an LCD screen or an e-ink screen. The temperature sensor is pre-installed inside the terminal device to detect the operating temperature of the internal components.

[0082] Those skilled in the art will understand that Figure 6 The schematic diagram shown is only a partial structural diagram related to the present invention and does not constitute a limitation on the terminal device to which the present invention is applied. The specific terminal device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0083] In one embodiment, a terminal device is provided, including a memory and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by one or more processors. The one or more programs include instructions for performing the following operations: Based on the geographical grid division of the construction section, a multimodal sensor cluster is deployed to collect real-time data on on-site personnel behavior, construction equipment, construction environment parameters, and the real-time status of construction entity quality, thereby obtaining multi-source heterogeneous real-time status data to construct a global perception layer. Edge-side preprocessing is performed on the collected multi-source heterogeneous real-time status data. The preprocessed data is then spatiotemporally and semantically aligned with the BIM model to generate digital twin mapping data, thereby constructing a data processing layer. The edge-side preprocessing includes: protocol parsing, outlier cleaning, and keyframe extraction. The digital twin mapping data is deeply analyzed based on a pre-trained AI model library to output structured decision results in order to build an intelligent engine layer. The AI ​​model library includes: a progress prediction model based on LSTM time series network, a quality defect identification model based on YOLOv8 visual network, or a multi-factor safety risk early warning model based on random forest algorithm. Configure a business rules engine to transform the structured decision results into standardized business control instructions, and distribute the business control instructions to the corresponding field control equipment or business subsystems to achieve closed-loop control of the construction process, thereby building a business application layer; A role-based access control mechanism is established to create a correlation index for data at each level, providing differentiated visual interactive interfaces and functional permissions for different participants, thereby constructing a user interaction layer.

[0084] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a railway engineering construction section-level platform based on temporal prediction and visual inspection networks, characterized in that, The method includes: Based on the geographical grid division of the construction section, a multimodal sensor cluster is deployed to collect real-time data on on-site personnel behavior, construction equipment, construction environment parameters, and the real-time status of construction entity quality, thereby obtaining multi-source heterogeneous real-time status data to construct a global perception layer. Edge-side preprocessing is performed on the collected multi-source heterogeneous real-time status data. The preprocessed data is then spatiotemporally and semantically aligned with the BIM model to generate digital twin mapping data, thereby constructing a data processing layer. The edge-side preprocessing includes: protocol parsing, outlier cleaning, and keyframe extraction. The digital twin mapping data is deeply analyzed based on a pre-trained AI model library to output structured decision results in order to build an intelligent engine layer. The AI ​​model library includes: a progress prediction model based on LSTM temporal network, a quality defect identification model based on YOLOv8 visual network, and a multi-factor safety risk early warning model based on random forest algorithm. Configure a business rules engine to transform the structured decision results into standardized business control instructions, and distribute the business control instructions to the corresponding field control equipment or business subsystems to achieve closed-loop control of the construction process, thereby building a business application layer; A role-based access control mechanism is established to create a correlation index for data at each level, providing differentiated visual interactive interfaces and functional permissions for different participants, thereby constructing a user interaction layer. The spatiotemporal semantic alignment of the preprocessed data with the BIM model includes: Microsecond-level time synchronization of multi-source heterogeneous real-time status data is achieved using the Network Time Protocol. A BIM and GIS fusion engine is constructed to map multi-source heterogeneous real-time status data containing geographic coordinate information to the three-dimensional spatial coordinate system of the BIM model; Establish a mapping table between sensor IDs and BIM components, and attach multi-source heterogeneous real-time status data (non-spatial attributes) to the attributes of the corresponding BIM components to complete data fusion; Based on the construction schedule of railway engineering, the binding logic between the sensor ID and BIM components is dynamically adjusted according to the time nodes of the construction task book through a soft mapping mechanism, so that the same physical sensor can be associated with the corresponding monitoring identity in different time windows. The outlier cleaning and keyframe extraction in the edge-side preprocessing specifically include: using a sliding window filtering algorithm to remove outlier noise points in multi-source heterogeneous real-time status data; using an image sharpness evaluation algorithm to filter video stream data in multi-source heterogeneous real-time status data, discarding video frames with sharpness below a preset threshold, extracting keyframes, and compressing the extracted keyframes. The global perception layer executes an adaptive sampling rate adjustment strategy based on data volatility. It assesses the dynamic intensity of the on-site physical characteristics by calculating the first derivative or variance of the perceived values ​​in real time, and dynamically switches the sampling frequency between preset high-frequency and low-frequency ranges according to the first derivative or variance of the perceived values.

2. The method for constructing a railway engineering construction section-level platform based on temporal prediction and visual inspection networks according to claim 1, characterized in that, The operation process of the multi-factor security risk early warning model based on the random forest algorithm includes: Define a risk assessment index system, which includes construction entity risk factors and construction equipment risk factors; The weight coefficients of construction entity risk factors and construction equipment risk factors were determined using the analytic hierarchy process (AHP). The normalized real-time monitoring values ​​of construction entity risk factors and construction equipment risk factors are input into the multi-factor safety risk early warning model for classification and reasoning, and the comprehensive risk level of the current construction area is output. Based on the comprehensive risk level and in accordance with the preset risk classification and response mechanism, a structured decision result containing the location of the risk source, the consequences of the risk, and disposal suggestions is automatically generated. Specifically, based on the comparison results of environmental parameters monitored by the global perception layer with preset sensitivity thresholds, the component defect results identified by the quality defect identification model, and / or the resource allocation suggestions generated by the progress prediction model, the input features and / or weight parameters in the multi-factor safety risk early warning model are corrected in real time.

3. The method for constructing a railway engineering construction section-level platform based on temporal prediction and visual inspection networks according to claim 1, characterized in that, The operation process of the progress prediction model based on LSTM time series network includes: Input historical construction log data, resource input data, and environmental weather data; The temporal dependency features of construction progress are extracted by using an LSTM temporal network to predict the amount of work to be completed within a preset time window in the future. The deviation between the predicted project completion amount and the planned project duration is calculated. If the deviation exceeds the threshold, a structured decision result is generated, including a schedule delay warning and resource allocation suggestions. The operation process also includes: introducing prior constraints based on the construction organization logic of railway engineering, and performing logical consistency verification on the predicted engineering completion quantity according to the logical relationship of the components in the BIM model, so as to eliminate prediction fluctuations that do not conform to the construction organization logic.

4. The method for constructing a railway engineering construction section-level platform based on temporal prediction and visual inspection networks according to claim 1, characterized in that, Distribute business control commands to the corresponding field control equipment or business subsystems to achieve closed-loop control of the construction process, including: A closed-loop feedback mechanism is constructed to track the execution status of the business control instructions and the results of on-site rectification in real time after the business control instructions are distributed to the corresponding field control equipment or business subsystems. The business application layer also has a built-in instruction arbitration mechanism based on priority weights, which is used to perform conflict assessment, collaborative optimization and execution reordering of the business control instructions according to a preset dynamic weight matrix of security, quality and schedule when there are logical conflicts in the instructions output by multiple AI models.

5. The method for constructing a railway engineering construction section-level platform based on temporal prediction and visual inspection networks according to claim 1, characterized in that, The global perception layer also includes the following features: The system executes a dynamic grid evolution logic based on the distribution of construction risks. By extracting multi-source heterogeneous real-time status data within each grid and calculating the grid risk entropy value, when the risk entropy value exceeds a preset stability threshold, the basic grid of the corresponding area is recursively subdivided into microgrids, and the acquisition parameters of the multimodal sensor cluster in the corresponding area are adjusted in conjunction.

6. A railway engineering construction section-level platform construction system based on temporal prediction and visual inspection networks, characterized in that, The system is used to implement the steps of the railway engineering construction section-level platform construction method based on temporal prediction and visual inspection network as described in any one of claims 1-5. The system includes: The global perception layer construction module is used to deploy a multimodal sensor cluster based on the geographical grid division of the construction section, and to collect real-time data on the behavior of on-site personnel, construction equipment, construction environment parameters and the real-time status of construction entity quality, so as to obtain multi-source heterogeneous real-time status data. The data processing layer construction module is used to perform edge-side preprocessing on the collected multi-source heterogeneous real-time status data, and to perform spatiotemporal semantic alignment between the preprocessed data and the BIM model to generate digital twin mapping data. The edge-side preprocessing includes: protocol parsing, outlier cleaning and keyframe extraction. The intelligent engine layer construction module is used to perform deep analysis on the digital twin mapping data based on the pre-trained AI model library and output structured decision results. The AI ​​model library includes: a progress prediction model based on LSTM temporal network, a quality defect identification model based on YOLOv8 visual network, and a multi-factor safety risk early warning model based on random forest algorithm. The business application layer construction module is used to configure the business rule engine, transform the structured decision results into standardized business control instructions, and distribute the business control instructions to the corresponding field control equipment or business subsystems to achieve closed-loop control of the construction process. The user interaction layer module is used for role-based access control mechanisms, establishing correlation indexes for data at each level, and providing differentiated visual interactive interfaces and functional permissions for different participants.

7. A terminal device, characterized in that, The terminal device includes a memory, a processor, and a railway engineering construction section-level platform construction program based on a time-series prediction and visual inspection network, which is stored in the memory and can run on the processor. When the processor executes the railway engineering construction section-level platform construction program based on a time-series prediction and visual inspection network, it implements the steps of the railway engineering construction section-level platform construction method based on a time-series prediction and visual inspection network as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a railway engineering construction section-level platform construction program based on a time-series prediction and visual inspection network. When the railway engineering construction section-level platform construction program based on a time-series prediction and visual inspection network is invoked on the computer-readable storage medium, it implements the steps of the railway engineering construction section-level platform construction method based on any one of claims 1-5.

Images