Construction engineering construction management method and system based on big data

Abstract

The application provides a construction engineering construction management method and system based on big data, relates to the technical field of construction management, and extracts features of physical entities in a construction site and organizes time trees through the Internet of Things and visual perception to obtain a weight-decayed construction node feature set; gradient field calculation and spatial clustering are performed on a fault loss function through the weight-decayed construction node feature set to obtain a sensitive point distribution map; state tracking and deviation detection are performed on historical and real-time trend paths through the sensitive point distribution map and the construction node feature set to obtain a trend path of a current construction state on a time tree; fault loss minimization simulation and multi-objective arbitration are performed on a candidate construction instruction set through the trend path and the sensitive point distribution map to obtain a standardized construction instruction set, and the application solves the technical problems of data islands, lagging risk identification, fuzzy root cause positioning and lack of basis for decision-making in construction management, and significantly improves the initiative and accuracy of construction management.

Description

Technical Field

[0001] This invention proposes a construction project management method and system based on big data, which relates to the field of construction management technology, specifically to the field of construction project management based on big data. Background Technology

[0002] Construction project management involves multiple dimensions such as schedule, cost, quality, and safety. Traditional methods mainly rely on manual inspections and periodic reports. However, data sources such as IoT sensors, video surveillance, BIM models, and schedule plans at construction sites are isolated, forming data silos that cannot support comprehensive cross-dimensional analysis. The value of historical data decreases over time, and construction techniques and personnel skills continuously change. Data from too far back may mislead current decisions, and traditional fixed-time-window methods struggle to effectively balance the contribution weight of recent and long-term data. Risk identification relies on human experience, only recording losses after a failure occurs, failing to identify potential high-risk areas in advance. When schedules are delayed or costs exceed limits, managers often struggle to trace back to the earliest specific process and time of the anomaly, resulting in a lack of targeted corrective measures. Faced with the contradiction between schedule delays and cost control, regulatory decisions are often made based on experience, lacking quantitative evidence. Summary of the Invention

[0003] This invention provides a construction project management method and system based on big data to solve the above-mentioned problems: The present invention proposes a construction project management method and system based on big data, wherein the method includes: S1. By using the Internet of Things and visual perception, feature extraction and time-tree organization of physical entities at the construction site are performed to obtain a set of construction node features with weight decay. S2. Calculate the gradient field and perform spatial clustering on the fault loss function using the feature set of construction nodes with weight decay to obtain the distribution map of sensitive points. S3. By using the distribution map of sensitive points and the feature set of construction nodes, the status of historical and real-time trend paths is tracked and deviations are detected to obtain the trend path of the current construction status on the time tree. S4. By using trend paths and sensitive point distribution maps, simulate the failure loss minimization and multi-objective arbitration of candidate construction instruction sets to obtain standardized construction instruction sets.

[0004] Furthermore, the system includes: The feature extraction module is used to extract features and organize them into a time tree for physical entities at the construction site through the Internet of Things and visual perception, so as to obtain a set of construction node features with weight decay. The sensitive distribution module is used to calculate the gradient field and perform spatial clustering on the fault loss function through the feature set of construction nodes with weight decay, so as to obtain the sensitive point distribution map. The trend path module is used to track the status and detect deviations of historical and real-time trend paths by using the distribution map of sensitive points and the feature set of construction nodes, so as to obtain the trend path of the current construction status on the time tree. The construction instruction module is used to simulate the failure loss minimization and multi-objective arbitration of candidate construction instruction sets through trend paths and sensitive point distribution maps, so as to obtain a standardized construction instruction set.

[0005] The beneficial effects of this invention are as follows: This method solves problems such as data silos leading to delayed decision-making, outdated and invalid historical data, reliance on manual experience for risk areas, inability to pinpoint the root cause of deviations from the plan, and lack of quantitative assessment of control measures. It achieves intelligent forgetting of historical data through time trees and weight decay, automatically adapting to changes in construction conditions. It bridges the gap from point-based perception to continuous risk assessment across the entire site through gradient fields and spatial clustering, allowing inspection resources to be prioritized for high-risk areas. It achieves precise location of deviation points through trend path tracking and reverse tracing, supporting targeted correction. It provides quantitative decision support through pre-execution simulation and multi-objective arbitration, improving decision-making quality. It forms a complete management closed loop of perception, analysis, decision-making, execution, and feedback through standardized instruction encapsulation. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of a construction project management method based on big data. Detailed Implementation

[0007] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0008] In one embodiment of the present invention, a construction project management method and system based on big data is proposed, wherein the method includes: S1. By using the Internet of Things and visual perception, feature extraction and time-tree organization of physical entities at the construction site are performed to obtain a set of construction node features with weight decay. S2. Calculate the gradient field and perform spatial clustering on the fault loss function using the feature set of construction nodes with weight decay to obtain the distribution map of sensitive points. S3. By using the distribution map of sensitive points and the feature set of construction nodes, the status of historical and real-time trend paths is tracked and deviations are detected to obtain the trend path of the current construction status on the time tree. S4. By using trend paths and sensitive point distribution maps, simulate the failure loss minimization and multi-objective arbitration of candidate construction instruction sets to obtain a standardized construction instruction set, such as... Figure 1 As shown.

[0009] The sensing devices are deployed at the construction site as hardware terminals, including RTK-GPS positioning base stations, UWB indoor positioning tags, MEMS triaxial accelerometers, strain gauges, temperature and humidity sensors, and industrial network cameras (visible light and infrared).

[0010] The original physical quantity signal is provided and converted from analog to digital to generate the original sensing data stream.

[0011] The device must have an industrial-grade protection rating (IP65 or higher), support edge data preprocessing (such as filtering and noise reduction), and prioritize MQTT or CoAP communication protocols to reduce power consumption.

[0012] Raw sensing data stream is the unprocessed binary or structured output of sensing devices, including timestamps, device IDs, numerical / image frames, and location coordinates (latitude, longitude, and elevation).

[0013] It serves as the lowest-level input for subsequent spatial matching and feature statistics.

[0014] A circular buffer needs to be designed to temporarily store the most recent N frames to prevent data loss due to network jitter; Protocol Buffers or MessagePack should be used for compact serialization.

[0015] Architectural design models (BIM models) are digital deliverables of engineering projects based on IFC or Revit formats, containing information such as component geometry, spatial coordinates, material properties, and construction process requirements.

[0016] It provides a standard reference system for matching the spatial location of construction components and binding engineering semantics.

[0017] The BIM coordinate system needs to be transformed into the on-site measurement control network (such as CGCS2000) using seven parameters to ensure that the matching error is ≤5cm.

[0018] Spatial location matching calculates the spatial topological relationship (such as point-to-surface containment and line-to-surface intersection) between the location coordinates in the sensing data and the bounding boxes of components in the BIM model, thus determining the specific component associated with the current data.

[0019] Establish a one-to-one correspondence between physical entities and digital models.

[0020] R-tree spatial indexing can be used to accelerate matching; for positioning drift caused by occlusion / multipath effects, Kalman filtering is required for smoothing correction.

[0021] Engineering semantic binding adds engineering business attributes to the matched components, such as process type (tying / pouring / curing), responsible team, design strength level, and planned construction period.

[0022] This gives spatial points business meaning, making it easier for subsequent multidimensional feature statistics.

[0023] The binding relationships are stored in a label-property graph, which supports SPARQL or Gremlin queries.

[0024] After spatial matching and semantic binding, each construction component is abstracted into a data point with spatial coordinates and engineering attributes. All data points constitute the discrete point set of the entire field.

[0025] It serves as the basic spatial unit for subsequent feature statistics and gradient calculation.

[0026] Each point in the point set needs to be assigned a globally unique ID (such as UUID) and associated with a parent component (such as a steel bar belonging to a beam segment).

[0027] Multidimensional feature statistics are used to calculate the progress completion rate, quality acceptance pass rate, machinery utilization rate, personnel attendance rate, and material consumption deviation for each construction stage according to time windows (such as day / shift), forming a high-dimensional vector.

[0028] Fragmented perception data is aggregated into computable features at the stage level.

[0029] The feature dimension is usually between 20 and 50, and Min-Max normalization or Z-score standardization is required to eliminate the influence of the dimension.

[0030] The overall timeline approved in the project's overall schedule construction organization design is usually expressed as a network diagram prepared using the Critical Path Method (CPM) or the Program Evaluation and Review Technique (PERT).

[0031] It provides theoretical time anchors for the construction process, serving as the time axis framework for the time tree organization.

[0032] It is necessary to import XML / MPP files exported from Project or P6 software to automatically parse the FS / SS / FF / SF logical relationships between processes.

[0033] The time-dependent decay rule is a calculation strategy that reduces the weight over time. The core formula is exponential decay w(t)=e-λ·Δt or half-life form w(t)=2-Δt / Tw, where Δt is the time difference between the record entry and the current time, and TT is the half-life parameter.

[0034] Control the rate at which the influence of historical data on the current state decays.

[0035] The value of TT needs to be differentiated according to the type of process. During the concrete strength development period (first 28 days), T=7 days, and during the later stage, T=60 days; for progress data, T=15 days. The attenuated weights need to be normalized so that the sum is 1 to avoid numerical deviation.

[0036] The process-adaptive attenuation rule set is a rule library that configures the half-life and attenuation curve type (exponential / power law / piecewise linear) for different construction processes (such as earthwork excavation, rebar installation, concrete pouring, masonry, and plastering).

[0037] This makes the weight decay more consistent with the aging physical characteristics of each process.

[0038] The rule set is stored in JSON Schema, supports dynamic hot reloading, and does not require restarting the service when changes are made.

[0039] The on-site fault record log is a database of various issues and events recorded daily by the project quality and safety department, including quality problems (honeycomb surface, cracks, dimensional deviations), safety hazards (lack of edge protection, improper use of electricity), schedule deviations (days of delay), and cost overruns (rework costs).

[0040] Supervisory labels are used to construct the fault loss function.

[0041] The ledger needs to be stored in a structured format, with each record associated with a specific construction site ID and timestamp, and marked with a severity level (general / major / critical).

[0042] Multidimensional loss accounting converts each failure event into a unified loss dimension (usually in RMB or man-hours), including direct losses (material scrap, wasted machine hours) and indirect losses (delay penalties, credit score deductions).

[0043] Quantifying fault events into continuous values ​​facilitates function fitting.

[0044] The calculation of indirect losses should take into account the contract breach clauses and the company's internal risk reserve system, and the calculation factor should be jointly reviewed and approved by the legal and financial departments.

[0045] The gradient field of the loss change across the entire field is calculated at each spatial point (x,y) on the field. The partial derivative vector ∇L=(∂L / ∂x,∂L / ∂y) of the loss function LL with respect to the location is calculated, forming a vector field covering the entire field (the gradient magnitude can be taken for the scalar field).

[0046] It reveals the sensitive direction and intensity of loss as it changes with spatial location.

[0047] Gradient calculation should use the central difference scheme, and forward / backward difference should be used at the boundaries; regions where the gradient magnitude exceeds the threshold (such as twice the standard deviation of the historical mean) are high-sensitivity regions.

[0048] Spatial extrapolation methods (interpolation algorithms) are a set of methods that use the measured gradient values ​​of a finite number of discrete sampling points to extrapolate the gradient values ​​of areas without sensors through spatial interpolation. Commonly used methods include Kriging interpolation or inverse distance weighting (IDW).

[0049] Solve the problem of blind spots in perception caused by insufficient equipment deployment density.

[0050] Kriging interpolation requires a semi-variogram fitting (spherical / exponential / Gaussian model); the power parameter p of IDW is usually taken as 2~4. Before interpolation, inaccessible areas formed by site obstacles (such as tower crane foundations, material storage yards) need to be deducted, and CDT (constrained Delaunay triangulation) is used to limit the calculation range.

[0051] The standard error of the interpolation results for each estimation point of the confidence distribution is calculated using cross-validation (such as leave-one-out method), and the reciprocal of the error is used as the confidence index, with a value range of [0,1].

[0052] This provides a quantitative reference for the reliability of data in subsequent decision-making, avoiding over-reliance on areas of high uncertainty.

[0053] For areas with a confidence level below 0.6, it is recommended to deploy additional mobile sensing devices (such as drone inspections).

[0054] The intensity of resource competition, which is the competition among multiple construction stages for limited resources (tower crane usage time, concrete pump trucks, skilled workers) within the same time window, is calculated as the ratio of resource application volume to available resource volume.

[0055] Used to determine whether adjacent sensitive points in space should be merged into the same high-risk area (high competition will amplify losses).

[0056] By employing DBSCAN or OPTICS clustering algorithms and using resource competition as a weighting factor for cluster distance, region merging can be achieved.

[0057] The historical state sequence is formed by traversing the time tree from root to leaf, and arranging the state vectors of each node at a historical moment (including four types of indicators: progress, quality, safety, and resources) in chronological order.

[0058] Provide time series samples for trend forecasting.

[0059] The sequence length is fixed as a sliding window (e.g., the most recent 30 time steps), and missing values ​​are filled using linear interpolation or forward padding.

[0060] The predictive ability of construction status change trends is achieved by using a time-series prediction model (such as LSTM, Transformer, or XGBoost) to learn the mapping function s^t and 1=f(st,st-1,...,st-k) from historical state sequences.

[0061] The core engine for rolling simulations of future construction status.

[0062] Model training requires the use of TimeSeriesSplit to avoid future information leakage; hyperparameters (number of hidden layers, learning rate) are automatically tuned through Bayesian optimization.

[0063] The expected state trajectory is a curve derived from the project plan showing how the theoretical state changes over time, including the percentage of planned progress at each time point, the planned quality acceptance points, and the planned resource input curve.

[0064] As a benchmark for comparison with the actual state trajectory.

[0065] The planned logical relationships (CPM) need to be transformed into time-state discrete points, with the sampling interval consistent with the historical state sequence (e.g., daily).

[0066] The trajectory deviation index measures the distance between the actual trajectory and the expected trajectory at the same point in time. It is usually expressed as a weighted Euclidean distance or a dynamic time warping (DTW) distance, with the weights determined by the partial derivatives of the fault loss function with respect to each state dimension.

[0067] The severity of quantitative deviation.

[0068] DTW requires setting window constraints (such as Sakoe-Chiba bands) to avoid excessive distortion; if the deviation exceeds a preset threshold (such as 15%), an out-of-limit flag will be triggered.

[0069] State change point detection identifies statistically significant jumps in state over a time series using change point detection algorithms (such as PELT and Binary Segmentation).

[0070] Quickly locate deviations from the starting point during backtracking of out-of-standard data.

[0071] During detection, a minimum segment length (no less than 3 sampling points) should be set to avoid false alarms due to noise; the confidence level of mutation should be evaluated using Bayesian factor analysis.

[0072] The pre-set control measures template library contains pre-compiled templates for common construction risk scenarios. Each template includes the measure type (adjusting the construction sequence / increasing resources / modifying process parameters / strengthening monitoring), adjustable parameters (such as adding 2-3 vibrators or slowing down the pouring speed to 0.8m / h), and applicable conditions (risk level, type of work surface).

[0073] Digitizing expert experience accelerates the generation of candidate instructions.

[0074] The template library is organized using a decision tree or rule engine (such as Drools), supports version management, and provides feedback on the confidence level of the updated rule after each successful application.

[0075] The candidate control instruction set is a set of several specific executable instructions generated by matching the current risk spatiotemporal window from the template library and substituting in real-time on-site parameters (such as the current idle time of the tower crane and the arrival time of the concrete mixer truck).

[0076] This provides alternative options for optimizing the screening process.

[0077] Each instruction must include execution prerequisites (such as "two pump trucks need to be available simultaneously"). After generation, a rapid pre-check of the feasibility of the conditions must be performed to eliminate unmet requirements.

[0078] The post-execution state deduction treats candidate instructions as external intervention actions, substitutes them into the trend prediction model, and simulates the future state trajectory (including time and space dimensions) after the instruction is executed from the current moment.

[0079] Predict the actual effect of each instruction and calculate the expected change in loss.

[0080] The simulation needs to take into account the lag effect of the command (e.g., after adjusting the pouring speed, the strength change will not be apparent until 12 hours later), and a time series model with external input (e.g., SARIMAX or Transformer with exogenous variables) should be used.

[0081] The difference between the expected loss of the entire cycle after executing the candidate instruction (Lafter) and the expected loss of the original path (Lbefore) without executing the instruction (ΔL=Lbefore-Lafte), and then the resource consumption cost (Ccost) of the instruction execution itself, is deducted to obtain the net benefit (ΔLnet=ΔL-Ccost).

[0082] It serves as the core quantitative indicator for ranking the quality of instructions.

[0083] Resource consumption costs must include direct costs (additional labor / machinery costs) and opportunity costs (losses from occupying resources in other processes), the latter being calculated using a resource competition model.

[0084] Pareto optimization selects the set of all candidate instructions that are not simultaneously dominant by other instructions in all objectives within a multi-objective space (time, quality, safety, cost), which is called the Pareto front.

[0085] Compress the candidate set to extract effective compromise solutions from multi-objective conflicts.

[0086] The algorithm is implemented using NSGA-II or MOEA / D, with the number of generations set to 100-200, and the population size is the same as the number of candidate instructions.

[0087] The standardized construction instruction set is the optimal instruction finally selected after multi-objective arbitration, and is a structured instruction list packaged in a unified format (including instruction number, issuance time, execution object, operation content, safety briefing attachments, expected effect, and validity period).

[0088] It is issued to the on-site execution end and serves as a reference benchmark for subsequent closed-loop feedback.

[0089] The instruction set must be accompanied by a digital signature (such as the SM2 national cryptographic algorithm) to ensure that it cannot be tampered with; at the same time, a human-readable natural language version should be generated for the team leader to sign.

[0090] The working principle and technical effects of the above technical solution are as follows: This invention collects the location and status data of personnel, machinery, and materials on-site through the Internet of Things and visual perception. Combined with time-tree organization, the data is mounted onto a multi-level time hierarchy, and each historical data point is assigned a weight that decays exponentially over time, outputting a construction node feature set with weight decay. Based on this feature set and historical fault records, a fault loss function is established. The gradient field of loss as a function of location is calculated on the construction point set. Spatial extrapolation completes the entire field data, high-risk areas are screened and spatial clustering is performed, outputting a sensitive point distribution map. Combining the sensitive point distribution map and the construction node feature set, a historical state sequence is constructed, and the trend prediction capability is obtained by learning the state change patterns. During real-time operation, the future state trajectory is continuously extrapolated, compared with the planned trajectory to obtain the degree of deviation. When deviations exceed the limit, the deviation point is traced back along the time tree to locate the starting point and output the actual trend path. Based on the actual trend path, identify the high-risk time and space window that is about to be entered, enumerate candidate control instructions from the preset template library, simulate the execution of each instruction and calculate the expected loss reduction, screen the preferred instructions through multi-objective Pareto comparison, select the optimal instruction according to the on-site management priority, attach credibility assessment and decision reasons, and output a standardized construction instruction set to push to the on-site execution terminal.

[0091] This method addresses issues such as data silos leading to delayed decision-making, outdated and invalid historical data, reliance on manual experience for risk areas, inability to pinpoint the root causes of deviations from plans, and lack of quantitative assessment for control measures. It achieves intelligent forgetting of historical data through time-based trees and weight decay, automatically adapting to changes in construction conditions. Gradient fields and spatial clustering enable a leap from point-based perception to continuous risk assessment across the entire site, allowing inspection resources to be prioritized for high-risk areas. Trend path tracking and reverse tracing enable precise location of deviation points, supporting targeted correction. Pre-execution simulation and multi-objective arbitration provide quantitative decision support, improving decision quality. Standardized instruction encapsulation forms a complete management closed loop encompassing perception, analysis, decision-making, execution, and feedback.

[0092] In one embodiment of the present invention, S1 includes: Real-time spatial positioning and identity recognition of the construction scene are performed by sensing devices to obtain raw sensing data streams; By matching the spatial location of construction components and machinery with engineering semantics through the raw perception data stream and the architectural design model, a set of construction points is obtained. By using the construction point set and historical project archives, multi-dimensional feature statistics are performed on each construction stage to obtain the initial construction stage feature vector. By organizing all construction stages in a time hierarchy using the initial construction stage feature vectors and the overall project schedule, a construction stage feature set under a time tree structure is obtained. By using the construction stage feature set under the time tree structure and the timeliness decay rule, the influence degree of each historical record is weighted and calculated to decrease over time, thus obtaining the construction stage feature set with weight decay.

[0093] The working principle and technical effects of the above-mentioned technical solution are as follows: This method uses millimeter-wave radar, structured light cameras, and other sensing devices deployed on tower cranes, fences, and key passages to collect real-time location information and identification of personnel, machinery, and material storage areas, forming a raw sensing data stream with timestamps and spatial coordinates. Subsequently, the system spatially matches the raw sensing data stream with the BIM design model, associating the sensed entities with the nearest BIM components through coordinate transformation and inheriting their engineering semantics. Each matched component or piece of machinery becomes a construction point, containing information such as location coordinates, positioning uncertainty range, and associated process number. The system utilizes the set of construction points and historical project archives to perform multi-dimensional feature statistics on each construction stage, including planned duration, actual duration reference value, required types and quantities of work, resource competition index, and historical failure frequency, forming an initial construction stage feature vector. Based on the overall project schedule, the system organizes all construction stages into a time tree according to time hierarchy. The root node represents the total project duration, and each stage is refined to the hourly level as leaf nodes. Each leaf node is associated with the feature vector of the corresponding construction stage, forming a construction stage feature set under the time tree structure. The system introduces a time-sensitive decay rule, which calculates the exponential decay weight based on the time interval between the entry time of each data and the current time, so that recent data has a large weight and long-term data has a small weight. After weighted reorganization, the system outputs a feature set of construction stages with weight decay.

[0094] This method addresses the challenges of unifying the organization of multi-source heterogeneous data and decision-making biases caused by outdated historical data. Through intelligent binding of perception and BIM, it achieves digital twin mapping between physical entities and engineering components, enabling precise component-level analysis. Multi-dimensional feature statistics aggregate information scattered across schedule, cost, and quality systems into a unified feature space, providing one-stop status viewing. Time-tree organization enables cross-scale data retrieval capabilities, meeting the viewing needs of managers at different levels. A weight decay mechanism allows the system to automatically forget outdated information, adapting to changing construction conditions without manual intervention and avoiding the difficulty of determining window length in fixed-window methods.

[0095] In one embodiment of the present invention, the step of performing a weighted calculation on the influence degree of each historical record by applying a construction stage feature set under a time tree structure and a timeliness decay rule, with the weight decreasing over time, to obtain a construction stage feature set with weighted decay, includes: By using the construction process feature set under the time tree structure and the entry time of historical records, the time difference of each historical data is calculated to obtain the time interval of each data from the current moment. By calculating the weight coefficient of each historical data point based on the time interval between each data point and the current time and the preset decay half-life parameter, the individual data weights that decrease exponentially over time are obtained. By differentiating the decay half-life by individual data weights and different construction process types, a decay rule set adapted to the process is obtained. By weighting and recombining the attenuation rule set adapted to the process and the feature vector of the construction process, the updated feature value of the historical and current data is obtained. By updating feature values ​​and time tree node indexes, all construction stages are uniformly stored and version-marked to obtain a set of construction stage features with weight decay.

[0096] Specifically, a weighting coefficient is calculated for each historical data point by using the time interval between each data point and the current time and a preset decay half-life parameter, to obtain the individual data weight that decreases exponentially over time, including: The decay rate constant is calculated by substituting the time interval between each data point and the current time with the preset decay half-life parameter into the formula, and the decay rate coefficient applicable to the data is obtained. The exponential part of the exponential function is numerically calculated by multiplying the decay rate coefficient by the negative value of the time interval, and the intermediate calculation result of the exponent power is obtained. The original, uncorrected weighting coefficients are obtained by calculating the value of the exponential function using the intermediate calculation result of the exponent and the natural constant e. The original weights are adjusted by amplifying or reducing their importance based on the importance level of the construction process (critical process, ordinary process, auxiliary process), thus obtaining the corrected weight coefficients that are suitable for each process. By normalizing and balancing the corrected weight coefficient with the sum of the weights of all historical data in the same stage (ensuring that the sum of all weights in the same stage is a fixed value), we obtain the individual data weights that decrease exponentially over time and are normalized.

[0097] The working principle and technical effects of the above technical solution are as follows: This method obtains the entry time of each historical data and the current system time, calculates the time difference between the two, and the calculation accuracy of the time difference can be selected at the hourly or daily level according to the construction management needs. The system introduces a preset decay half-life parameter, which can be configured according to the characteristics of different construction stages. The decay rate coefficient is calculated through a formula, which is inversely proportional to the half-life. The time difference is multiplied by the decay rate coefficient to obtain an intermediate product value. The system uses the natural constant as the base and the negative value of the product as the exponent to calculate the exponential function value, obtaining the uncorrected original weight coefficient, whose physical meaning is that the data weight decays smoothly over time. The system performs importance correction on construction stages of different importance levels. The correction coefficient for key processes is greater than the benchmark value to slow down the decay rate, while the correction coefficient for auxiliary processes is less than the benchmark value to accelerate the decay. The corrected weight is equal to the original weight multiplied by the importance correction coefficient. The system performs normalization and balancing on all weights within the same construction phase. First, it calculates the sum of the corrected weights of all historical data in that phase, and then divides the corrected weight of each data point by the sum to make the sum of weights within the same phase a fixed value. This eliminates the impact of uneven data sampling density on weighted statistics and outputs individual data weights that decrease exponentially over time and are normalized.

[0098] This method addresses the shortcomings of traditional weighted methods, such as lack of theoretical basis, inconsistent treatment of processes of varying importance, and statistical bias caused by uneven sampling density. The exponential decay mechanism ensures that the rate of weight decrease is proportional to the current weight value, achieving a rapid initial decay followed by a slower decay, which aligns with the high sensitivity of construction management to recent data. The half-life parameter provides an intuitive and controllable tool for adjusting the time scale, allowing managers to adapt to different scenarios without needing to understand complex mathematics. Importance correction ensures that historical data of key processes retains its influence for a longer period, aligning the system's focus with the actual importance of the project. Normalization eliminates statistical bias caused by changes in sampling frequency, ensuring stable system output and preventing unnecessary fluctuations due to changes in data recording habits.

[0099] In one embodiment of the present invention, S2 includes: By using the construction phase feature set with weighted decay and the on-site fault record ledger, multi-dimensional loss accounting is performed on each type of fault phenomenon to obtain the fault loss function. By calculating the sensitivity of loss to location changes at each construction point using the fault loss function and the spatial location of the construction point set, the gradient field of loss change for the entire site is obtained. By using the loss change gradient field of the entire field and spatial extrapolation method, the gradient values ​​of the area where sensing devices are not directly deployed are spatially extrapolated to obtain a loss gradient distribution map that continuously covers the entire field. By using a loss gradient distribution map that continuously covers the entire field and risk assessment criteria, high-risk areas are screened for all spatial locations to obtain a candidate set of sensitive points. By combining the candidate set of sensitive points with the degree of resource competition in the construction process characteristics, spatially adjacent sensitive points are merged and divided into regions to obtain a sensitive point distribution map.

[0100] The working principle and technical effects of the above technical solution are as follows: This method associates the feature set of construction stages with weight decay with the on-site fault record ledger, classifies and statistically analyzes direct economic losses, construction delay penalties, safety compensation, reputational damage, etc., according to fault type, and establishes a parameterized fault loss function. This function can output a comprehensive loss estimate based on fault type, location, and duration. The system calculates the partial derivative of the loss function with respect to spatial coordinates on the set of construction points, uses the numerical difference method to make small offsets in six directions at each construction point and calculates the rate of change of loss, synthesizes the gradient vector, and obtains the gradient field of loss change on discrete sampling points. For areas where no sensing devices are deployed, the system estimates the gradient value through spatial extrapolation: it searches for the nearest known sample points around the unknown point, assigns weights according to the inverse of the distance, and checks the directional distribution of the sample points for balance correction to avoid the estimation result being biased towards the direction of sample density. The extrapolated gradient value is obtained by weighted averaging. The system sets a gradient magnitude threshold as a risk judgment criterion and selects locations with gradients exceeding the threshold as a candidate set of sensitive points. The system merges adjacent sensitive points in space, taking the degree of resource competition as the merging criterion. It groups adjacent sensitive points that share the same tower crane or other resources into the same sensitive area and outputs a sensitive point distribution map. Each area is accompanied by information such as the average risk level, a list of processes involved, and the recommended inspection frequency.

[0101] This method addresses several issues in traditional safety management. By employing a failure loss function, it achieves unified quantification of risks of different natures, such as quality, safety, schedule, and cost, on the same scale, enabling priority comparison of cross-category risks. Through a loss gradient field, it upgrades the approach from post-failure statistics to pre-failure risk prediction, identifying potentially high-risk areas that, while not yet experiencing failures, could result in significant losses. Spatial extrapolation solves the challenge of continuous risk assessment across the entire site under sparse sensing equipment deployment, providing managers with a risk view of any location on the construction site. Sensitive point clustering integrates fragmented risk points into manageable risk areas, improving the operational efficiency and intuitiveness of inspection plan development.

[0102] In one embodiment of the present invention, the step of spatially extrapolating the gradient values ​​of areas where sensing devices are not directly deployed by using a loss variation gradient field and spatial extrapolation method across the entire field to obtain a loss gradient distribution map continuously covering the entire field includes: By extracting and verifying the gradient values ​​at discrete sampling points using the loss change gradient field across the entire site and the location points of the deployed sensing devices, a discrete gradient sample set is obtained. By defining the passable area of ​​the spatial extrapolation using a discrete gradient sample set and the distribution of site space boundaries and obstacles, a constrained spatial extrapolation range is obtained. By using the spatial distance and directional relationship between the constrained spatial extrapolation range and the known gradient sample points, the gradient values ​​of unknown locations are weighted and estimated to obtain the extrapolated gradient values ​​of each unmarked point. The confidence level of the inference results is evaluated by comparing the inferred gradient values ​​of each unmarked point with the measured gradient values ​​of the original sampling points, thereby obtaining the gradient values ​​and confidence level distribution of all points in the entire site. By continuously coloring and smoothing the transition of the entire field area using gradient values ​​and confidence distributions and an electronic site map, a loss gradient distribution map is obtained.

[0103] Specifically, the gradient values ​​of unknown locations are weighted and estimated by the spatial distance and directional relationship between the constrained spatial extrapolation range and the known gradient sample points, to obtain the extrapolated gradient values ​​for each unmarked point, including: By using the constrained spatial extrapolation range and the spatial coordinates of known gradient sample points, the nearest neighbor sample points are searched for each unknown location point to obtain the nearest known sample points to the unknown point. The spatial distance weight of each sample point is calculated by taking the inverse of the distance between the nearest known sample point and the spatial distance from each sample point to the unknown point (the closer the distance, the greater the weight). The weight distribution is corrected for directional balance by using the spatial distance weight of the sample points and the mutual directional angle between the sample points (to avoid all high-weight samples being concentrated in the same direction), thus obtaining spatially isotropic corrected weights. By adjusting the weights and calculating the weighted average with the measured gradient values ​​of each known sample point, a preliminary estimated gradient value for the unknown location point can be obtained. By adjusting the preliminary gradient values ​​with geological coefficients based on the site's geological conditions (soft soil area, hard soil area, and fill area), the final calculated gradient values ​​for each unmarked point are obtained.

[0104] The working principle and technical effect of the above solution are as follows: For each unknown location point, the system searches for the nearest known gradient sample points in its vicinity. If the number of sample points within the search radius is insufficient, the radius is gradually expanded to ensure that each unknown point has sufficient reference. The system calculates the reciprocal of the distance from each sample point to the unknown point as the initial weight, with a larger weight for closer points. The system performs directional balance correction on the weight distribution, dividing the space around the unknown point into several angular sectors, and summing the weights of each sector. If the weight of a certain sector is significantly too high, the weight of the sample points in that sector is compressed, and the weights of other sectors are increased accordingly to make the directional distribution of the weights more uniform, avoiding deviations in the estimation results due to sample points being concentrated in one direction. The system calculates a weighted average of the corrected weights and the measured gradient values ​​of the sample points to obtain the preliminary estimated gradient value of the unknown location point. The system adjusts the geological coefficients according to the site's geological conditions, dividing the construction site into geological units such as natural foundation area, backfill soil area, and soft soil area. Each unit has a preset adjustment coefficient. The gradient value is initially calculated and multiplied by the geological coefficient of the location to obtain the final calculated gradient value, so that the estimation results reflect the spatial heterogeneity of the geological conditions.

[0105] This method addresses issues such as uneven sample distribution leading to estimation bias and the failure to incorporate geological condition differences into risk assessment during spatial extrapolation. The inverse distance weighting method is computationally efficient, suitable for real-time operation of edge devices, and conforms to the first law of geography. Directional balance correction resolves the uneven sample distribution caused by the deployment of sensing devices along linear structures, ensuring that estimation results are referential and usable in all directions. Geological coefficient adjustment integrates engineering geological knowledge into risk assessment, automatically increasing gradient estimates in soft soil areas and appropriately decreasing them in rocky areas, thus improving the physical accuracy of gradient distribution. Ultimately, this achieves quantifiable estimation of gradient values ​​at every location on the construction site.

[0106] In one embodiment of the present invention, S3 includes: By using the sensitive point distribution map and the construction stage feature set with weight decay, the construction status features of the smallest unit on each time tree are extracted to obtain the historical state sequence. By learning patterns from historical state sequences and time-dependent decay rules, the ability to predict the trend of construction state changes can be obtained. By using the ability to predict the trend of changes in construction status and real-time sensing data streams, the current construction status is continuously simulated to obtain the predicted status trajectory. By quantitatively comparing the state differences at the same point in time between the predicted state trajectory and the expected state trajectory derived from the project plan, an index of trajectory deviation can be obtained. The current construction status is judged to exceed the standard by using the trajectory deviation index and the allowable deviation range. If the standard is exceeded, the deviation point is traced back to obtain the actual trend path of the current construction status on the time tree.

[0107] The working principle and technical effects of the above technical solution are as follows: This method combines a sensitive point distribution map with a construction stage feature set based on weight decay. It extracts a snapshot of the construction status from each smallest unit on the time tree, containing information such as progress percentage, resource consumption, quality indicators, safety status, and whether it is located in a sensitive area. These snapshots are then arranged chronologically to obtain a historical state sequence. The system uses the historical state sequence and time-sensitivity decay rules to learn the patterns of state changes over time, establishing a state transition model. This model can predict future evolution trends based on the state sequence over a past period, with recent states having higher weights in the learning process. The resulting predictive ability reflects the current construction mode. During real-time operation, the system continuously receives sensing data, constructs the current state sequence, and inputs it into the state transition model. It then continuously extrapolates the predicted state values ​​for multiple future time points, forming a predicted state trajectory. The system compares the predicted state trajectory with the expected state trajectory derived from the project plan at the same time point, comprehensively considering deviations across multiple dimensions such as progress differences, resource differences, and quality differences, and calculates the trajectory deviation index according to preset weights. If the deviation index exceeds the allowable range, the system traces backward from the current node to the root node along the time tree, detects the state change rate of each ancestor node, finds the mutation point where the change rate suddenly increases, locates the earliest mutation point that conforms to causal logic as the deviation starting point, continuously reconstructs the actual state sequence from the starting point to the current moment, and outputs the actual trend path of the current construction state on the time tree.

[0108] This method addresses several issues in traditional management. Through time-series pattern learning, it achieves forward-looking capabilities in the construction process, enabling managers to intervene before problems escalate, shifting from reactive response to proactive prevention. The rolling projection mechanism continuously incorporates the latest real-time data into the forecast results, automatically correcting previous deviations and ensuring that forecasts are always synchronized with actual site conditions. Multi-dimensional deviation indicators provide objective quantitative assessments, avoiding the bias of subjective judgment. The reverse tracing mechanism enables precise identification of the root causes of deviations, allowing managers to find the cause at the specific process level, rather than being at a loss when facing the entire project.

[0109] In one embodiment of the present invention, the step of judging whether the current construction status exceeds the standard by using the trajectory deviation index and the allowable deviation range, and if the standard is exceeded, tracing back to find the starting point of the deviation to obtain the actual trend path of the current construction status on the time tree, includes: The deviation of each time tree node is determined by comparing the trajectory deviation index with the preset allowable deviation range, and the sequence of nodes that exceed the allowable deviation is obtained. By traversing the parent-child relationships of out-of-standard nodes through the out-of-standard node label sequence and the hierarchical structure of the time tree, the complete ancestor path of the out-of-standard nodes can be obtained. By using the complete ancestor path of the out-of-standard node and the state change records of each node, a state mutation point detection is performed on each ancestor node to obtain a candidate set of deviations from the starting point. By verifying the causal rationality of the deviation starting point through the logical dependency relationship between the candidate deviation starting point set and the construction stage, the earliest deviation starting point is obtained. By continuously reconstructing the actual construction path on the timeline using the earliest deviation from the starting point and the state snapshot sequence from the starting point to the current moment, the state and deviation of each node are marked to obtain the actual trend path.

[0110] Specifically, by analyzing the complete ancestor paths of the out-of-standard nodes and the state change records of each node, a state mutation point detection is performed on each ancestor node to obtain a candidate set of deviations from the starting point, including: By calculating the state difference between adjacent time points for each ancestor node using the complete ancestor path of the out-of-standard node and the state change records of each node on the time axis, the state change rate sequence of each node is obtained. The current rate is significantly larger by comparing the state change rate sequence of each node with the change rate range of the node's historical normal state (statistically derived from the historical state sequence) to determine whether the current rate is significantly larger, and thus obtain the rate anomaly marker for each node. By using rate anomaly markers and the duration of the anomaly (the rate is high at multiple consecutive time points), transient fluctuations and continuous mutations are distinguished and screened to obtain a candidate set of nodes with continuous state mutations. The candidate nodes are sorted on a timeline by the candidate set of nodes with continuous state mutations and the order in which the mutations occur. Starting from the earliest node, the subsequent nodes are checked to see if they mutate synchronously, thus obtaining the classification of independent mutation points and chained mutation points. By classifying independent and linked mutation points and ranking all mutation points by the magnitude of their state changes (the size of the transition from normal to abnormal values), the root cause probability of each mutation point is determined (the mutation point with the largest magnitude and the most independent is most likely to be the deviation from the starting point), thus obtaining a set of candidate deviation starting points.

[0111] The working principle and technical effects of the above solution are as follows: For each node on the ancestral path of the out-of-standard node, the system extracts its state change records, calculates the state difference between adjacent time points divided by the time interval, and obtains a state change rate sequence. The system compares the current rate with the rate range obtained from historical normal periods. If the current rate significantly deviates from the historical normal range (e.g., exceeding the mean plus or minus several times the standard deviation), the system marks that moment as a rate anomaly. The historical benchmark can be dynamically updated due to the use of a weighted decay feature set of construction stages, adapting to the gradual change of construction mode. The system checks the duration of the anomaly marking. If the anomaly duration is less than the preset short-term fluctuation threshold, it is considered random noise and ignored. If it exceeds the continuous mutation threshold, it is marked as a continuous state mutation, avoiding false alarms for minor disturbances in normal construction. The system sorts the selected continuous mutation nodes by occurrence time and checks whether subsequent mutations may be caused by the earliest mutation. By analyzing the historical influence relationship between state variables, the system distinguishes between independent mutation sources and chain reaction nodes. The system sorts independent mutation nodes according to the magnitude of state change (the size of the transition from normal to abnormal value). The larger the transition magnitude, the more serious the impact of the node on subsequent construction, and therefore the more likely it is to be the true deviation point. The system outputs a set of candidate deviation points sorted by root cause probability.

[0112] This method addresses the technical challenges of high false alarm rates and ambiguous root cause localization in state mutation detection. Through dynamic comparison with historical normal rate ranges, it achieves objective quantitative judgment of abnormal states, avoiding the uncertainty of subjective threshold settings. The duration-based screening mechanism effectively distinguishes between random fluctuations and genuine mutations, significantly reducing the system's false alarm rate. The technique for differentiating independent mutation sources from chain reactions enables the system to identify the true root cause node rather than the triggered symptom node, which is crucial for accurate root cause localization. Sorting by mutation magnitude allows managers to prioritize the most severely affected deviation starting point, improving problem-solving efficiency. Overall, this method elevates deviation localization from a general description of a problem occurring in a certain area to a precise description of a specific node, a specific time, and a specific magnitude of deviation.

[0113] In one embodiment of the present invention, S4 includes: By combining the actual trend path of the current construction status on the time tree with the distribution map of sensitive points, a dual overlay analysis of time and space is conducted on the high-risk areas that are about to be entered, to obtain a list of risk spatiotemporal windows. By listing all possible control actions and filling in parameters through a risk spatiotemporal window list and a preset control measure template library, a set of candidate control instructions is obtained. By combining the candidate control command set with the ability to predict the trend of construction status changes, the state of each candidate command is deduced after execution. The expected loss after execution is calculated by substituting it into the fault loss function, and the amount of expected loss reduced by the command is obtained. By comparing the reduction of expected loss by instructions and the multidimensional impact attributes of candidate instructions, a superior set of instructions is obtained by hierarchically comparing the merits of all candidate instructions. By optimizing the instruction set and using on-site management priority rules, the optimal instruction is finally selected and its reliability is evaluated to obtain a standardized set of control instructions.

[0114] The working principle and technical effects of the above technical solution are as follows: Based on the actual trend path and sensitive point distribution map, this method identifies high-risk areas that construction is about to enter. Through time and space superposition analysis, it determines the expected arrival time and expected stay duration of each high-risk area, forming a risk spatiotemporal window list. The system enumerates all feasible control actions from the preset control measure template library, including resource allocation, process adjustment, quality rectification, safety warning, etc., and fills in specific parameters (such as adjustment object, change value, execution location) for each instruction, forming a candidate control instruction set. The system uses trend prediction capabilities to simulate the execution of each candidate instruction, substitutes the instruction effect into the state transition model, recalculates the future state trajectory, and substitutes it into the failure loss function to obtain the expected total loss after execution. Subtracting this from the expected loss of the original path yields the expected loss reduction. The system evaluates the comprehensive impact of each instruction on multi-dimensional objectives (cost, schedule, safety), and selects the optimal instruction set through Pareto comparison, i.e., those instructions that cannot improve one objective without worsening another. The system selects the optimal instruction from the preferred set based on the on-site management priority rules (such as prioritizing progress when the penalty for delay exceeds the cost of catching up), attaches a credibility assessment and explanation of the decision reasons, and outputs a standardized set of control instructions to the on-site execution terminal.

[0115] This method addresses technical issues such as the lack of quantitative basis for regulatory decisions and the inability to balance conflicting multiple objectives. The generation of a risk spatiotemporal window list allows managers to anticipate upcoming high-risk periods and areas, providing a window of opportunity for decision-making. Pre-execution simulation technology enables quantitative prediction of the effects of each candidate instruction, allowing managers to clearly see how much expected loss each measure can reduce. The Pareto comparison method rationally handles the conflicting objectives of cost, schedule, and safety, avoiding contradictory instructions output by the system. The final selected optimal instruction is accompanied by credibility and rationale, enhancing human-machine trust and making managers more willing to adopt the system's recommendations.

[0116] In one embodiment of the present invention, the step of performing post-execution state deduction for each candidate instruction by using a set of candidate control instructions and the ability to predict the trend of construction state changes, substituting the results into a fault loss function to calculate the expected loss after execution, and obtaining the amount by which the instruction reduces the expected loss, includes: By comparing the set of candidate control commands with the actual trend path of the current construction status, the execution conditions of each candidate command are checked to obtain a subset of candidate commands. By using a subset of candidate instructions and the ability to predict the trend of changes in construction status, the post-execution status of each candidate instruction is deduced to obtain the future status trajectory. By recalculating the spatiotemporal window of crossing high-risk areas after instruction execution using future state trajectories and sensitive point distribution maps, the risk exposure situation after instruction execution can be obtained. The expected total loss after execution is calculated step by step by using the risk exposure and failure loss function, and the difference is compared with the expected loss of the original path to obtain the expected loss reduction brought by each instruction. The net expected loss reduction for each candidate instruction is obtained by adjusting the net benefit of the reduction based on the expected loss reduction from the resource consumption and time cost required for instruction execution.

[0117] Specifically, by using a subset of candidate instructions and the ability to predict trends in construction status changes, the post-execution state of each candidate instruction is extrapolated to obtain its future state trajectory, including: By extracting the baseline state of each candidate instruction at the time of its effective date using a subset of candidate instructions and the actual trend path of the current construction status, a snapshot of the state before the instruction is executed is obtained. By directly modifying the resource dimension values ​​in the state snapshot before the command is executed and the resource adjustment parameters (increase or decrease in the number of personnel, machinery, and materials) contained in the command, the state after resource adjustment at the moment the command takes effect can be obtained. By calculating the effect of resource changes on process progress through the logical dependency relationship between the state after resource adjustment and the process (the impact coefficient of acceleration or deceleration of subsequent processes after resources are in place), the progress correction amount of each affected process can be obtained. By using the progress correction amount and the prediction capability of construction status change trend (the original trend prediction model), the corrected complete state sequence is rolled out (the correction amount is substituted into the prediction as a new initial condition) to obtain the state prediction value at each future time point after the execution of the instruction. By continuously arranging and aligning the predicted values ​​with the time tree node index after the command is executed, the future state trajectory starting from the time the command takes effect is obtained (each time point includes the status of each construction point and the characteristics of each link).

[0118] The working principle and technical effects of the above technical solution are as follows: For each candidate instruction, the system captures a snapshot of the state at the moment the instruction takes effect in the actual trend path of the current construction state, which serves as the baseline state for deduction. Based on the resource adjustment parameters included in the instruction (such as the increase or decrease in the number of personnel, machinery, and materials), the system directly modifies the resource dimension values ​​in the baseline state snapshot to obtain the resource-adjusted state at the moment the instruction takes effect. Based on the logical dependencies between processes, the system deduces the transmission effect of resource changes on the progress of subsequent processes. For example, adding a concrete pump truck will accelerate the pouring process, thereby advancing the subsequent curing and formwork removal processes accordingly. The system calculates the progress correction amount for each affected process. The system uses the progress correction amount as a new initial condition, substitutes it into the original state transition model for rolling deduction, simulates the complete state evolution process starting from the moment the instruction takes effect, and obtains the predicted state values ​​at each future time point. The system arranges the predicted values ​​continuously in chronological order and aligns them with the time tree node index, ensuring that the state description at each time point includes the state of each construction point and the characteristics of each construction stage at that moment, and outputs the complete future state trajectory starting from the moment the instruction takes effect.

[0119] This method addresses technical challenges such as the difficulty in quantifying and predicting the effects of instructions, and incompatibility between simulations and existing prediction models. By capturing baseline states and directly modifying resource dimensions, it achieves clear and traceable tracking of instruction effects. The calculation of process logic dependencies and progress transmission establishes a quantitative link between local resource adjustments and their global impact on progress, avoiding the one-sidedness of isolated assessments of resource changes. The technical solution of substituting correction values ​​as new initial conditions into the original prediction model ensures consistency between the simulation process and the system's original trend prediction capabilities, making the simulation results reliable.

[0120] According to one embodiment of the present invention, the system includes: The feature extraction module is used to extract features and organize them into a time tree for physical entities at the construction site through the Internet of Things and visual perception, so as to obtain a set of construction node features with weight decay. The sensitive distribution module is used to calculate the gradient field and perform spatial clustering on the fault loss function through the feature set of construction nodes with weight decay, so as to obtain the sensitive point distribution map. The trend path module is used to track the status and detect deviations of historical and real-time trend paths by using the distribution map of sensitive points and the feature set of construction nodes, so as to obtain the trend path of the current construction status on the time tree. The construction instruction module is used to simulate the failure loss minimization and multi-objective arbitration of candidate construction instruction sets through trend paths and sensitive point distribution maps, so as to obtain a standardized construction instruction set.

[0121] The working principle and technical effects of the above technical solution are as follows: This invention collects the location and status data of personnel, machinery, and materials on-site through the Internet of Things and visual perception. Combined with time-tree organization, the data is mounted onto a multi-level time hierarchy, and each historical data point is assigned a weight that decays exponentially over time, outputting a construction node feature set with weight decay. Based on this feature set and historical fault records, a fault loss function is established. The gradient field of loss as a function of location is calculated on the construction point set. Spatial extrapolation completes the entire field data, high-risk areas are screened and spatial clustering is performed, outputting a sensitive point distribution map. Combining the sensitive point distribution map and the construction node feature set, a historical state sequence is constructed, and the trend prediction capability is obtained by learning the state change patterns. During real-time operation, the future state trajectory is continuously extrapolated, compared with the planned trajectory to obtain the degree of deviation. When deviations exceed the limit, the deviation point is traced back along the time tree to locate the starting point and output the actual trend path. Based on the actual trend path, identify the high-risk time and space window that is about to be entered, enumerate candidate control instructions from the preset template library, simulate the execution of each instruction and calculate the expected loss reduction, screen the preferred instructions through multi-objective Pareto comparison, select the optimal instruction according to the on-site management priority, attach credibility assessment and decision reasons, and output a standardized construction instruction set to push to the on-site execution terminal.

[0122] This system addresses issues such as data silos leading to delayed decision-making, outdated and invalid historical data, reliance on manual experience for risk areas, inability to pinpoint the root causes of deviations from plans, and lack of quantitative assessment for control measures. It achieves intelligent forgetting of historical data through time-based trees and weight decay, automatically adapting to changes in construction conditions. Gradient fields and spatial clustering enable a leap from point-based perception to continuous risk assessment across the entire site, allowing inspection resources to be prioritized for high-risk areas. Trend path tracking and reverse tracing enable precise location of deviation points, supporting targeted correction. Pre-execution simulation and multi-objective arbitration provide quantitative decision support, improving decision-making quality. Standardized instruction encapsulation forms a complete management closed loop encompassing perception, analysis, decision-making, execution, and feedback.

[0123] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A construction project management method based on big data, characterized in that, The method includes: S1. By using the Internet of Things and visual perception, feature extraction and time-tree organization of physical entities at the construction site are performed to obtain a set of construction node features with weight decay. S2. Calculate the gradient field and perform spatial clustering on the fault loss function using the feature set of construction nodes with weight decay to obtain the distribution map of sensitive points. S3. By using the distribution map of sensitive points and the feature set of construction nodes, the status of historical and real-time trend paths is tracked and deviations are detected to obtain the trend path of the current construction status on the time tree. S4. By using trend paths and sensitive point distribution maps, simulate the failure loss minimization and multi-objective arbitration of candidate construction instruction sets to obtain standardized construction instruction sets.

2. The construction project management method based on big data according to claim 1, characterized in that, S1 includes: Real-time spatial positioning and identity recognition of the construction scene are performed by sensing devices to obtain raw sensing data streams; By matching the spatial location of construction components and machinery with engineering semantics through the raw perception data stream and the architectural design model, a set of construction points is obtained. By using the construction point set and historical project archives, multi-dimensional feature statistics are performed on each construction stage to obtain the initial construction stage feature vector. By organizing all construction stages in a time hierarchy using the initial construction stage feature vectors and the overall project schedule, a construction stage feature set under a time tree structure is obtained. By using the construction stage feature set under the time tree structure and the timeliness decay rule, the influence degree of each historical record is weighted and calculated to decrease over time, thus obtaining the construction stage feature set with weight decay.

3. The construction project management method based on big data according to claim 2, characterized in that, The method involves weighting the impact of each historical record under a time-tree structure using a construction stage feature set and a timeliness decay rule, with the weight decreasing over time, to obtain a construction stage feature set with weighted decay, including: By using the construction process feature set under the time tree structure and the entry time of historical records, the time difference of each historical data is calculated to obtain the time interval of each data from the current moment. By calculating the weight coefficient of each historical data point based on the time interval between each data point and the current time and the preset decay half-life parameter, the individual data weights that decrease exponentially over time are obtained. By differentiating the decay half-life by individual data weights and different construction process types, a decay rule set adapted to the process is obtained. By weighting and recombining the attenuation rule set adapted to the process and the feature vector of the construction process, the updated feature value of the historical and current data is obtained. By updating feature values ​​and time tree node indexes, all construction stages are uniformly stored and version-marked to obtain a feature set of construction stages with weight decay.

4. The construction project management method based on big data according to claim 1, characterized in that, S2 includes: By using the construction phase feature set with weighted decay and the on-site fault record ledger, multi-dimensional loss accounting is performed on each type of fault phenomenon to obtain the fault loss function. By calculating the sensitivity of loss to location changes at each construction point using the fault loss function and the spatial location of the construction point set, the gradient field of loss change for the entire site is obtained. By using the loss change gradient field of the entire field and spatial extrapolation method, the gradient values ​​of the area where sensing devices are not directly deployed are spatially extrapolated to obtain a loss gradient distribution map that continuously covers the entire field. By using a loss gradient distribution map that continuously covers the entire field and risk assessment criteria, high-risk areas are screened for all spatial locations to obtain a candidate set of sensitive points. By combining the candidate set of sensitive points with the degree of resource competition in the construction process characteristics, spatially adjacent sensitive points are merged and divided into regions to obtain a sensitive point distribution map.

5. The construction project management method based on big data according to claim 4, characterized in that, The method of spatially extrapolating gradient values ​​in areas where sensing devices are not directly deployed, using the loss change gradient field of the entire field and a spatial extrapolation method, yields a loss gradient distribution map that continuously covers the entire field, including: By extracting and verifying the gradient values ​​at discrete sampling points using the loss change gradient field across the entire site and the location points of the deployed sensing devices, a discrete gradient sample set is obtained. By defining the passable area of ​​the spatial extrapolation using a discrete gradient sample set and the distribution of site space boundaries and obstacles, a constrained spatial extrapolation range is obtained. By using the spatial distance and directional relationship between the constrained spatial extrapolation range and the known gradient sample points, the gradient values ​​of unknown locations are weighted and estimated to obtain the extrapolated gradient values ​​of each unmarked point. The confidence level of the inference results is evaluated by comparing the inferred gradient values ​​of each unmarked point with the measured gradient values ​​of the original sampling points, thereby obtaining the gradient values ​​and confidence level distribution of all points in the entire site. By continuously coloring and smoothing the transition of the entire field area using gradient values ​​and confidence distributions and an electronic site map, a loss gradient distribution map is obtained.

6. The construction project management method based on big data according to claim 1, characterized in that, S3 includes: By using the sensitive point distribution map and the construction stage feature set with weight decay, the construction status features of the smallest unit on each time tree are extracted to obtain the historical state sequence. By learning patterns from historical state sequences and time-dependent decay rules, the ability to predict the trend of construction state changes can be obtained. By using the ability to predict the trend of changes in construction status and real-time sensing data streams, the current construction status is continuously simulated to obtain the predicted status trajectory. By quantitatively comparing the state differences at the same point in time between the predicted state trajectory and the expected state trajectory derived from the project plan, an index of trajectory deviation can be obtained. The current construction status is judged to exceed the standard by using the trajectory deviation index and the allowable deviation range. If the standard is exceeded, the deviation point is traced back to obtain the actual trend path of the current construction status on the time tree.

7. The construction project management method based on big data according to claim 6, characterized in that, The process of judging whether the current construction status exceeds the standard by using the trajectory deviation index and the allowable deviation range, and if it does, tracing back to find the starting point of the deviation to obtain the actual trend path of the current construction status on the time tree, includes: The deviation of each time tree node is determined by comparing the trajectory deviation index with the preset allowable deviation range, and the sequence of nodes that exceed the allowable deviation is obtained. By traversing the parent-child relationships of out-of-standard nodes through the out-of-standard node label sequence and the hierarchical structure of the time tree, the complete ancestor path of the out-of-standard nodes can be obtained. By using the complete ancestor path of the out-of-standard node and the state change records of each node, a state mutation point detection is performed on each ancestor node to obtain a candidate set of deviations from the starting point. By verifying the causal rationality of the deviation starting point through the logical dependency relationship between the candidate deviation starting point set and the construction stage, the earliest deviation starting point is obtained. By continuously reconstructing the actual construction path on the timeline using the earliest deviation from the starting point and the state snapshot sequence from the starting point to the current moment, the state and deviation of each node are marked to obtain the actual trend path.

8. The construction project management method based on big data according to claim 1, characterized in that, S4 includes: By combining the actual trend path of the current construction status on the time tree with the distribution map of sensitive points, a dual overlay analysis of time and space is conducted on the high-risk areas that are about to be entered, to obtain a list of risk spatiotemporal windows. By listing all possible control actions and filling in parameters through a risk spatiotemporal window list and a preset control measure template library, a set of candidate control instructions is obtained. By combining the candidate control command set with the ability to predict the trend of construction status changes, the state of each candidate command is deduced after execution. The expected loss after execution is calculated by substituting it into the fault loss function, and the amount of expected loss reduced by the command is obtained. By comparing the reduction of expected loss by instructions and the multidimensional impact attributes of candidate instructions, a superior set of instructions is obtained by hierarchically comparing the merits of all candidate instructions. By optimizing the instruction set and using on-site management priority rules, the optimal instruction is finally selected and its reliability is evaluated to obtain a standardized set of control instructions.

9. The construction project management method based on big data according to claim 8, characterized in that, The process of using a set of candidate control commands and the ability to predict the trend of changes in construction status to perform post-execution state deduction for each candidate command, substituting it into a fault loss function to calculate the expected loss after execution, and obtaining the amount of expected loss reduction by the command includes: By comparing the set of candidate control commands with the actual trend path of the current construction status, the execution conditions of each candidate command are checked to obtain a subset of candidate commands. By using a subset of candidate instructions and the ability to predict the trend of changes in construction status, the post-execution status of each candidate instruction is deduced to obtain the future status trajectory. By recalculating the spatiotemporal window of crossing high-risk areas after instruction execution using future state trajectories and sensitive point distribution maps, the risk exposure situation after instruction execution can be obtained. The expected total loss after execution is calculated step by step by using the risk exposure and failure loss function, and the difference is compared with the expected loss of the original path to obtain the expected loss reduction brought by each instruction. The net expected loss reduction for each candidate instruction is obtained by adjusting the net benefit of the reduction based on the expected loss reduction from the resource consumption and time cost required for instruction execution.

10. A construction project management system based on big data, characterized in that: The system includes: The feature extraction module is used to extract features and organize them into a time tree for physical entities at the construction site through the Internet of Things and visual perception, so as to obtain a set of construction node features with weight decay. The sensitive distribution module is used to calculate the gradient field and perform spatial clustering on the fault loss function through the feature set of construction nodes with weight decay, so as to obtain the sensitive point distribution map. The trend path module is used to track the status and detect deviations of historical and real-time trend paths by using the distribution map of sensitive points and the feature set of construction nodes, so as to obtain the trend path of the current construction status on the time tree. The construction instruction module is used to simulate the failure loss minimization and multi-objective arbitration of candidate construction instruction sets through trend paths and sensitive point distribution maps, so as to obtain a standardized construction instruction set.

Images