Intelligent municipal engineering construction progress management methods

Abstract

This invention relates to the field of information technology for municipal engineering construction management, and discloses an intelligent method for managing the progress of municipal engineering construction. The method includes the following steps: A mobile terminal mounted on construction machinery collects data in real time and generates progress semantic events with spatiotemporal tags locally; the spatial occupancy projection of the machinery in the future time window is calculated in real time, generating an intent vector which is broadcast to neighboring machinery and uploaded to the cloud; the terminal receives intent vectors from other terminals, detects conflicts through spatiotemporal overlap, and if conflicts exist, autonomously avoids them according to preset negotiation rules and updates the intent vector until they are eliminated; the cloud aggregates progress events to construct a spatiotemporal graph, uses a graph neural network to learn the process pattern, and obtains efficiency analysis results; the cloud dynamically marks critical path resources based on the analysis results, issues priority tags, and the negotiation rules adjust the yielding order of each machine according to the priority tags, realizing rule-guided dynamic resource allocation. This invention can improve the coordination efficiency and progress control capabilities of municipal engineering construction.

Description

Technical Field

[0001] This invention relates to the field of information technology for municipal engineering construction management, and specifically to an intelligent method for managing the progress of municipal engineering construction. Background Technology

[0002] Municipal engineering projects, especially linear projects such as road paving and pipeline installation, are core components of urban infrastructure construction. The level of their construction progress management directly affects the total project cost, on-time performance, and the impact on urban traffic and residents' lives. With the continuous advancement of urbanization, municipal engineering projects are exhibiting significant characteristics such as increased scale, tighter schedules, and greater environmental sensitivity. Traditional construction progress management relies heavily on the project manager's experience and judgment, paper or electronic reports submitted by on-site personnel, and periodic progress meetings. This management model has significant shortcomings in terms of the timeliness of data collection, the depth of analysis, and the dynamism of decision-making.

[0003] In recent years, with the rapid evolution of IoT sensing technology, high-precision positioning technology, and artificial intelligence algorithms, the automation and intelligent management of construction progress through technological means has become an inevitable trend in the industry. Its core objective is to shift from passive post-event correction to proactive pre-event prediction and real-time adjustment, thereby minimizing project delays and their resulting negative consequences. At the current level of construction progress perception, existing technical solutions attempt to use GPS locators and visual sensors to track and monitor construction machinery and work areas. This has, to some extent, enabled the digital collection and remote uploading of progress data, and initially constructed a visual dashboard for progress management. Simultaneously, some management platforms have begun to introduce machine learning algorithms, training models based on historical progress data to assess current construction efficiency and attempt to reveal bottleneck factors affecting progress.

[0004] However, the applicability of the aforementioned technical solutions faces significant limitations in the unique scenarios of linear municipal engineering projects. Firstly, the work surface of linear engineering projects is constantly moving, making it difficult for fixed-deployment visual sensors to achieve effective coverage and continuous monitoring over construction zones several kilometers long, compromising the continuity and integrity of data collection. Furthermore, GPS signals are easily blocked in areas such as overpasses and tunnels, leading to data gaps and positioning blind spots. Secondly, practice shows that dozens of machines with different functions, such as excavators, dump trucks, pavers, and road rollers, operating serially or in parallel within a strip-shaped space only a few meters wide, are highly susceptible to conflicts and deadlocks due to encroachment on turning radii, occupation of the only access route, or overlapping entry and exit times. If effective avoidance and coordination cannot be provided before conflicts occur, it will result in a large number of unplanned downtimes and idle work surfaces, representing the most hidden and costly source of efficiency loss at municipal construction sites. Existing analytical methods struggle to trace the root causes of such delays from a holistic perspective. Third, scheduling decisions are highly time-sensitive: in a construction site where dozens of machines are operating in parallel, conflicts can occur within minutes, requiring a high degree of timeliness. Summary of the Invention

[0005] The present invention aims to provide an intelligent method for managing the construction progress of municipal engineering projects, which can significantly improve the coordination efficiency and progress control capabilities of municipal engineering construction without increasing the investment in a large amount of fixed infrastructure.

[0006] The basic solution provided by this invention is: an intelligent municipal engineering construction progress management method, including the following steps: Step 1: The mobile terminal mounted on each construction machine collects the positioning data, visual data and machine condition data of the construction machine in real time as the construction machine moves along the linear work surface, and generates progress semantic events with spatiotemporal tags locally on the mobile terminal. Step 2: Based on the current process status and mechanical motion parameters, each mobile terminal calculates the spatial occupancy projection of the construction machinery in a future preset time window in real time, generates an intent vector, and broadcasts it to the neighboring construction machinery through the direct communication link between the mobile terminals. At the same time, the progress semantic event and intent vector are uploaded to the cloud platform. Step 3: The first mechanical terminal receives the intent vector of at least one second mechanical terminal and determines whether there is a predictive conflict through spatiotemporal overlap detection. If there is, the terminals autonomously make avoidance decisions according to preset negotiation rules and update their respective intent vectors until the conflict is eliminated. The first mechanical terminal refers to a mobile terminal that acts as the active execution end of conflict detection, and the second mechanical terminal refers to another mobile terminal that acts as the end that receives and analyzes the intent vector. Step 4: The cloud platform gathers the progress semantic events of the entire line, constructs a spatiotemporal diagram of the construction process, uses graph neural networks to learn the normal process connection pattern, identifies abnormal progress nodes and attributes them to the spatiotemporal conflict events of the machine group that caused the abnormality, and outputs the efficiency analysis results. Step 5: Based on the efficiency analysis results, the cloud platform dynamically marks the critical path process resources that affect the overall project duration and sends priority tags to the corresponding mobile terminals; the negotiation rules adjust the yielding order of each construction machine in conflict negotiation according to the priority tags, so as to realize rule-guided dynamic resource allocation.

[0007] The working principle and advantages of this invention are as follows: This invention provides an intelligent method for managing the construction progress of municipal engineering projects. It can significantly improve the coordination efficiency and progress control capabilities of municipal engineering construction without increasing substantial investment in fixed infrastructure. The key points are: Unlike traditional methods that rely on fixed-point sensing or periodic manual reporting, this solution deploys sensing capabilities on mobile machinery, enabling data collection to flow synchronously with the work surface. This fundamentally solves the problem of discontinuous progress data caused by the continuous movement of sensing objects in linear engineering.

[0008] At the coordination mechanism level, this solution abandons the traditional delayed cycle of data collection-centralized analysis-instruction issuance. Instead, it enables direct exchange of future space occupancy intentions between construction machinery terminals, completing autonomous avoidance negotiations within a minute-level time window before actual conflicts occur. This mechanism transforms the micro-coordination process, which previously relied on operators' personal experience and spontaneous communication, into a calculable, recordable, and optimizable structured process, significantly reducing unplanned downtime and waiting caused by space encroachment or passageway blockage.

[0009] At the macro-analysis level, this solution does not simply count completed quantities or compare them with planned values. Instead, it constructs a spatiotemporal graph of the progress semantics and interaction relationships of the entire fleet. Using a graph attention network, it locates nodes with abnormal efficiency from a global structural perspective and traces upstream conflict events causing delays back along the related edges in the graph. This attribution approach ensures that the solution's output is not merely a superficial warning of "a certain delay," but rather points to the root cause of specific fleet conflicts, providing precise quantitative evidence for subsequent scheduling interventions. Based on this, the cloud platform dynamically identifies critical paths and adjusts priority rules in distributed negotiation accordingly, achieving a closed-loop linkage from macro-efficiency analysis to micro-behavioral guidance. Crucially, in this solution, the cloud platform does not directly schedule each machine. Instead, it dynamically calculates and distributes priority tags, altering the game rules of end-side negotiation. This lightweight approach guides the overall behavior of the fleet towards a globally optimal solution, avoiding the real-time bottleneck of centralized scheduling and overcoming the limitations of purely distributed negotiation that may fall into local optima. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating an embodiment of the intelligent municipal engineering construction progress management method of the present invention. Figure 2 This is a schematic diagram of the method flow of Embodiment 2 of the intelligent municipal engineering construction progress management method of the present invention. Detailed Implementation

[0011] The following detailed explanation illustrates the specific implementation methods: Example 1 The basic implementation examples are as follows: Figure 1 The following steps are shown in the intelligent municipal engineering construction progress management method: Step 1: The mobile terminal mounted on each construction machine collects the positioning data, visual data and machine condition data of the construction machine in real time as the construction machine moves along the linear work surface, and generates progress semantic events with spatiotemporal tags locally on the mobile terminal.

[0012] Specifically, the positioning data is obtained by fusing RTK-GPS with ultra-wideband UWB positioning anchors deployed along the work zone to maintain centimeter-level continuous positioning in areas where GPS signals are blocked; the visual data is collected by an industrial camera facing the work mechanism, and the machine condition data is obtained through a CAN bus or an external inertial measurement unit.

[0013] In this embodiment, the construction machinery includes pavers, road rollers, excavators, dump trucks, etc.

[0014] Each construction machine is equipped with a vehicle-mounted mobile terminal, which includes the following modules: (1) Positioning Module: Integrates an RTK-GPS receiver and a UWB tag. RTK-GPS achieves real-time positioning with a horizontal accuracy of ±2cm and an vertical accuracy of ±3cm in open areas by receiving differential correction data from a base station. UWB positioning anchors are deployed every 200 meters along the work area. The anchors emit ultra-wideband pulse signals. The mobile terminal UWB tag calculates the distance to the anchors using a time difference of arrival algorithm. In areas where GPS signals are blocked (such as under overpasses or in tunnels), it automatically switches to the primary positioning source and performs Kalman filtering fusion with the inertial measurement unit to maintain centimeter-level continuous positioning output. The positioning data update frequency is 10Hz, and each positioning record contains the fields {timestamp, longitude, latitude, elevation, positioning quality factor, positioning source identifier}.

[0015] (2) Vision module: A monocular industrial camera is selected and installed on the side of the machine operation direction. Among them, the paver camera faces the road surface in front of the hopper, the roller camera looks down at the roller track, and the excavator camera faces the bucket-loading interaction area.

[0016] (3) Machine condition sensing interface: Read mechanical operating parameters via CAN bus, including engine speed, travel speed, hydraulic system pressure, and lifting / tilting status. For older machines without a CAN interface, an external six-axis inertial measurement unit is fixed to the machine body to obtain three-axis acceleration and angular velocity, which are used to calculate the machine's motion state and working posture.

[0017] The progress semantic events are generated by reasoning from visual data using a lightweight convolutional neural network built into the mobile terminal, including at least: identifying the state of the material pile in front of the paver and the paving distance, identifying the texture changes associated with the number of compaction passes by the road roller, and identifying the number of loading cycles of the excavator; the progress semantic events are associated with station information to form a progress profile that is continuously distributed along the linear project.

[0018] Specifically, the resulting progress semantic events include: (1) Paving progress event: Identify the leading edge boundary of the asphalt mixture pile in the image collected by the vision module. When the cumulative forward advance distance of the pile reaches a preset threshold (e.g., 1 meter), a progress semantic event is generated by combining the positioning data of that period: {timestamp, machine ID:, station start and end, process type, progress event type, completed amount, quality parameter}; where the calculation method of station start and end is: the current positioning coordinates of the construction machine are perpendicularly projected to the center line of the project line to obtain the corresponding station K, and the current completed amount is accumulated to the starting station K0 to obtain the ending station K0+L, i.e., the format "K0~K0+L".

[0019] For example, {timestamp: "2026-05-07 14:23:15", machine ID: "P-02", chainage start and end: "K1+200~K1+201", process type: "asphalt paving", progress event type: "paving completed", completed quantity: 1 meter, quality parameter: average height of material pile 0.35 meters}.

[0020] (2) Compaction Pass Count Event: Identifies the texture features of the wheel tracks in the images acquired by the vision module. Asphalt pavement exhibits different surface texture patterns during the initial compaction, intermediate compaction, and final compaction stages. The network outputs the current texture category at a cycle of 0.5 seconds. When a complete compaction pass (a single pass of the steel wheel at the same point is counted as one pass) is identified, a progress semantic event is generated: {timestamp, machine ID, station number, process type, progress event type, current pass count, design pass count}.

[0021] For example, {timestamp, machine ID, station number, process type: "compaction", progress event type: "compaction pass update", current pass count: 3, design pass count: 5}.

[0022] (3) Excavation and Loading Cycle Events: Based on images acquired by the vision module, the sequence of actions of the bucket, namely "excavation-lifting-slewing-unloading-slewing reset", is identified. The peak pressure of the hydraulic system is combined with the visual determination of the unloading moment to confirm the end of a complete loading cycle. Each cycle generates a progress semantic event: {timestamp, machine ID, work station number, process type, progress event type, loading vehicle number, cycle time} For example, {timestamp, machine ID, work station number, operation type: "earthwork excavation and loading", progress event type: "loading cycle completed", loading vehicle number: "dump truck T-05", cycle time: 28 seconds}.

[0023] After the aforementioned progress semantic events are generated locally on the mobile terminal, they are sent to the cloud platform on the one hand, and used as the time reference input for subsequent intent calculation on the other hand.

[0024] Step 2: Based on the current process status, mechanical motion parameters, and mechanical geometric model, each mobile terminal calculates the spatial occupancy projection of the construction machinery in a future preset time window in real time, generates an intent vector, and broadcasts it to nearby construction machinery through the direct communication link between mobile terminals. At the same time, the progress semantic event and intent vector are uploaded to the cloud platform.

[0025] The intent vector includes at least: machine identifier, machine type, space occupancy area described by polygons or trajectories, time window in which the space occupancy area is effective, and the initial priority of the current operation of the construction machine.

[0026] Specifically, the preset time window length is set according to the work process type: 3 minutes for excavation and loading, 2 minutes for paving and compaction, and 1 minute for transportation. The spatial occupancy projection of each construction machine is calculated as follows: (1) Excavator space occupancy projection: The space occupancy projection within this time window is defined by taking the current slewing center coordinates of the excavator as the origin, the maximum operating radius R (e.g., 8 meters) as the radius, and the sector area formed by the current slewing start angle θ1 to the planned end angle θ2. The sector angle range depends on the geometric relationship between the current work surface position and the dump truck parking position. The projection is represented as the coordinate set of the polygon vertex sequence {P1 (slewing center), P2, P3, ...} in the engineering coordinate system.

[0027] (2) Projection of space occupied by the paver: Taking the front edge of the current hopper of the paver as the starting point, multiply the paving speed v (e.g., 3 m / min) by the length of the time window T to obtain the forward distance D = v × T. The projection of space occupied is a strip in front of the machine along the direction of travel, with a width equal to the paving width W (e.g., 6 m) and a length equal to D.

[0028] (3) Roller space occupancy projection: The calculation method is similar to that of the paver. The occupancy projection is a rectangular strip with a length equal to the rolling speed multiplied by the time window and a width equal to the width of the rolling wheel. If the current pass is not completed, the occupancy projection includes the reciprocating path required to complete the full pass of this section.

[0029] (4) Projection of space occupation of dump truck: including two parts - the driving path occupation from the current location to the loading point, and the static occupation area when parking and waiting at the loading point (safe distance of 1.5 meters outside the vehicle body outline).

[0030] Furthermore, the aforementioned space occupancy projection and metadata are assembled into an intent vector message, with the format: {machine ID, machine type, timestamp, occupied area, effective time window, current process, initial priority}.

[0031] For example, {Machine ID: "P-02", Machine type: "Paver", Timestamp: "2026-05-07 14:23:15", Occupied area: Polygon vertex coordinate set, Effective time window: [T_begin: "14:23:15", T_end: "14:25:15"], Current process: "Asphalt paving", Initial priority: 3}.

[0032] In this embodiment, the initial priority value is preset as follows: continuous shaping process such as paving and compaction is 1 (highest), key process of pipeline installation is 2, excavation and loading process is 3, and material transportation process is 4 (lowest).

[0033] Each mobile terminal broadcasts its own intent vector to nearby construction machinery within its communication range at a frequency of 10Hz via a direct communication link. The communication radius is approximately 300 meters, sufficient to cover machinery in the same work section and adjacent sections. Simultaneously, the intent vector and progress semantic events are forwarded to the cloud platform.

[0034] Step 3: The first mechanical terminal receives the intent vector from at least one second mechanical terminal and determines whether there is a predictive conflict through spatiotemporal overlap detection. If there is, the terminals autonomously make avoidance decisions based on preset negotiation rules and update their respective intent vectors until the conflict is eliminated. The first mechanical terminal refers to a mobile terminal that acts as the active execution end of conflict detection, and the second mechanical terminal refers to another mobile terminal that acts as the end that receives and analyzes the intent vector.

[0035] Specifically, the spatiotemporal overlap detection simultaneously determines the geometric intersection in space and the overlap in time windows; the predictive conflict classification is divided into hard conflicts of physical collision and soft conflicts of spatial encroachment that cause operation to be hindered.

[0036] In this embodiment, after receiving the intent vector broadcast by the neighboring second mechanical terminal, the first mechanical terminal performs collision detection, including the following sub-steps: (1) Spatial overlap judgment: Perform geometric intersection operation between the occupied area polygon in the local intent vector and the occupied area polygon in the received intent vector. Use a polygon clipping algorithm in computational geometry (such as the Sutherland-Hodgman algorithm) to calculate the area of ​​the intersection polygon. .like If so, then there is spatial overlap.

[0037] (2) Time overlap judgment: Compare the effective time windows of the two intent vectors. Local time window Adjacent machine time window If both conditions are met... and If so, the time windows overlap.

[0038] (3) Conflict confirmation: When both spatial and temporal overlaps are established, a predictive spatiotemporal conflict is confirmed.

[0039] (4) Conflict classification: Based on the nature of the occupied area according to the type of machinery, the conflict is divided into: hard conflict: two occupied polygons intersect directly and either involves mechanical rotation or invisible blind area, with a significant risk of physical collision; soft conflict: although there is no collision, the occupied area of ​​one party covers the necessary working space or only passage of the other party, causing the other party's process to be unable to be executed.

[0040] Furthermore, based on the judgment result, a conflict alarm record is generated: {Conflict ID, Conflicting Party} Conflict parties The remaining decision time is the time difference between the current moment and the expected time of conflict.

[0041] Once a conflict is confirmed, the terminals of both parties automatically enter a negotiation process, which includes the following sub-steps: (1) Information exchange: The conflicting parties exchange binary pairs (current priority, avoidance cost).

[0042] The priority acquisition method is as follows: the initial priority is inherited from the intent vector; if a dynamic priority label has been issued in the cloud, the latter will override the former.

[0043] The avoidance cost is calculated as follows: each terminal calculates the time loss incurred when it performs an active avoidance action. .

[0044] Specifically, for transportation machinery: = Increased travel time due to the avoidance route + potential queuing time (estimated based on the number of queuing machines at nearby loading / unloading points). For operational machinery: = The inherent time spent restarting after a pause (it takes about 15 seconds for the excavator to reposition the material pile, and the risk of cold joints caused by the paver paving pause is equivalent to about 30 seconds of time loss) + the time delay in the completion of this section caused by avoidance.

[0045] (2) Arbitration is conducted in accordance with the negotiation rules. In this embodiment, the negotiation rules are a rule-cost hybrid model based on priority and local avoidance cost: the conflicting parties exchange priorities and the time loss cost of avoidance actions calculated separately, and arbitration is completed based on the logic that the lower priority party gives way and the party with higher cost at the same priority has the right to demand that the lower cost party give way.

[0046] That is, if the two sides have different priorities, the one with the lower priority value (i.e., higher priority) gets the right-of-way, and the other side yields. If both parties have equal priorities, compare the costs of avoiding conflict between them. The machine with the lower obstacle avoidance cost will yield; if the obstacle avoidance costs are also equal, the machine with the smaller lexicographical order of its machine ID will yield by default.

[0047] (3) After the arbitration result is determined, the mobile terminal of the party making the decision shall recalculate its adjusted space occupancy projection and intent vector (such as adjusting the arrival time or changing the parking position), and broadcast the updated intent vector again. Then, a verification operation (re-detecting the conflict) shall be performed. If the conflict is eliminated, the agreement is reached; if the conflict still exists, the second negotiation shall be initiated or the cloud platform shall be reported to request manual intervention.

[0048] Step 4: The cloud platform aggregates the progress semantic events of the entire line, constructs a spatiotemporal diagram of the construction process, uses graph neural networks to learn the normal process connection pattern, identifies abnormal progress nodes and attributes them to the spatiotemporal conflict events of the machine group that caused the abnormality, and outputs the efficiency analysis results.

[0049] Specifically, the construction process spatiotemporal diagram uses each work segment as a node and the machine group movement path as an edge, embedding the progress status and machine group interaction features of each time period; the graph neural network is a graph attention network, which learns the spatiotemporal pattern of normal process connection, and when an abnormal progress stagnation node occurs, it outputs the upstream machine group conflict event that caused the stagnation and its impact range through attention weight backtracking.

[0050] In this embodiment, the cloud platform receives progress semantic event streams and intent vector streams from all terminals across the entire line in real time, aggregates them according to 5-minute time windows, and constructs a spatiotemporal diagram of the construction process. .

[0051] (1) Node V definition: The current project is divided into approximately 60 work segment nodes, with the smallest work segment as the node. Each node... (The f-th dimension feature component extracted from the i-th smallest process segment node) Take the feature vector within a 5-minute window, which includes: Static attributes: design process type, planned starting station number, planned construction period; Dynamic progress characteristics: current cumulative completion percentage (cumulative completion amount of generated progress events / design workload), changes in completion amount within the window; Resource occupancy characteristics: The distribution of the number and type of machinery serving this work section within the window.

[0052] (2) Definition of edge E: Process logic edge: connects adjacent work segment nodes of the preceding and following processes; Mechanical flow edge: If the same machine serves two work segments successively within the window, then connect the nodes of these two work segments; Cluster interaction edge: If two different machines have had overlapping or conflicting events in the same window, an interaction edge is established between the work segment nodes that each machine was serving or passing through at that time. The edge weight is the number of conflicts or the duration of overlap within the window.

[0053] In this embodiment, the construction, training, and application process of the graph neural network includes the following operations: (1) Model construction: A graph attention network is adopted, which contains three layers of multi-attention head graph convolutional layers. The input layer receives the multi-dimensional feature vectors of each node; the intermediate hidden layer maps the node features to a 64-dimensional embedding space; the output layer outputs the efficiency prediction value and anomaly score of each node in the next time window.

[0054] (2) Training data: The training set was constructed from the progress event logs of similar historical municipal engineering projects. Normally completed work segments were used as normal samples, and work segments that had experienced significant delays were used as abnormal samples. The network was trained to learn the spatiotemporal patterns of normal work segment connections. The loss function adopted was a weighted combination of cross-entropy and hinge loss to optimize the classification accuracy of node anomaly scores.

[0055] (3) Online inference: The model receives real-time spatiotemporal snapshots and outputs anomaly scores for each node. When the anomaly score of a node in a certain work segment exceeds a set threshold (e.g., 0.7), it is determined to be a node with abnormal progress and stagnation.

[0056] (4) Attention Attribution Backtracking: Backtracking is performed using the attention weight matrices stored in each layer of the graph attention network. Starting from the abnormal node, the attention weights of all upstream nodes connected to it in the entire link are extracted. Upstream nodes with attention weights higher than the average attention weight are marked as "high impact factor nodes".

[0057] Further examine the edge properties—if the edge between the high-impact node and the abnormal node is a "cluster interaction edge" and has a high edge weight (number of collisions / duration of overlap), the system outputs an attribution conclusion, for example: Abnormal Node: Pipeline installation at K1+200 (efficiency reduction of 60%) → Attribution Path: Earthwork excavation at K1+100 (attention weight 0.83) → Conflict Link: Excavator EX-03 and dump truck T-05 overlap in this section (conflict frequency: 3.2 times per cycle) → Impact: The average waiting time for dump trucks to enter the site increased from the baseline of 2 minutes to 15 minutes, affecting 3 downstream work sections.

[0058] Step 5: Based on the efficiency analysis results, the cloud platform dynamically marks the critical path process resources that affect the overall project duration and sends priority tags to the corresponding mobile terminals; the negotiation rules adjust the yielding order of each construction machine in conflict negotiation according to the priority tags, so as to realize rule-guided dynamic resource allocation.

[0059] Furthermore, the cloud platform continuously tracks the progress dependencies of each process segment and the delay propagation chain predicted by the graph neural network. Any process that will further delay the overall project duration is marked as a critical path process, and the mechanical resources serving that process are identified as critical resources. The priority label includes the critical resource label.

[0060] Specifically, in this embodiment, the cloud platform constructs an overall project schedule network using all process segments as nodes and process logic dependencies as edges. Based on the current remaining project duration prediction value of each node (calculated by combining the efficiency prediction value output by the graph neural network with the remaining project workload), the longest path of the entire project is calculated using the critical path method, which is the critical path. Delay at any node on the critical path will directly postpone the total project duration.

[0061] Furthermore, traverse all work segments on the critical path and extract a list of machinery resources currently serving and planned to serve the work segment (by matching the station number and work type in the intent vector). Identify these machinery resources as "critical resources" and generate priority labels: {Machine ID, Dynamic Priority, Effective Period, Issuance Time}.

[0062] For example, {Machine ID: "EX-03", Dynamic Priority: 1 (Super Critical), Effective Period: Immediately effective until completion of the current work segment, Issuance Time: "2026-05-07 14:30:00"}.

[0063] Mechanical maintenance or reset on non-critical paths is performed to the initial priority. Dynamic priority levels are added above the initial priority, including a supercritical level (level 0) and a subcritical level.

[0064] Then, through the communication link between the cloud platform and the mobile terminal, the priority tag is pushed to the corresponding mobile terminal, overriding the priority parameters in the terminal's local negotiation rules. After receiving the tag, the mobile terminal automatically participates in arbitration with the new priority in subsequent distributed conflict negotiation.

[0065] When the cloud platform identifies a systemic bottleneck that distributed negotiation cannot resolve—that is, when two or more processes marked as critical resources are competing for access—it generates a macro-scheduling command and directly sends it to the relevant mobile terminals. This command forcibly adjusts the construction sequence or spatial path of at least one of the processes, breaking the deadlock from a global perspective. This embodiment includes the following operations: (1) Impact assessment: Calculate the incremental impact of forced scheduling scheme A (making A yield) and scheme B (making B yield) on the total project duration. Using a graph neural network prediction model, quickly deduce the delay propagation chain of the two schemes in the next 72 hours in a digital thread environment, and output the total project duration delay.

[0066] (2) Generate instructions: Select the option with the smaller total project duration delay and generate a macro scheduling instruction: {Target machine ID, Instruction type, Content, Generation basis}. This instruction is directly sent to the target terminal, forcibly overriding the local negotiation result.

[0067] For example, {Target machine ID: "T-05", Instruction type: "Path replanning", Content: "Pause entry to K1+200 loading point, give way for 15 minutes, and enter after paver P-02 passes", Generation basis: "Critical path double conflict, estimated delay of 3.2 hours for scheme A (better than 4.8 hours for scheme B)"}.

[0068] In addition, for scheduling decisions involving transportation machinery such as dump trucks, the cloud platform also accesses dynamic traffic situation data from the urban road network surrounding the construction area.

[0069] In dynamic resource allocation, the cloud platform incorporates real-time traffic impedance of routes into the scheduling cost calculation when calculating scheduling strategies related to transportation machinery, generating a joint scheduling scheme that takes traffic impact into account. Specifically, this includes the following operations: (1) Traffic impedance calculation: Taking the current location of the dump truck as the starting point and the dump site as the ending point, the route planning API is called in real time to obtain the travel time of each alternative route. Congestion Index (CI) and Traffic Impedance ,in, The congestion penalty coefficient is calculated based on an empirical value of 5 minutes per congestion level.

[0070] (2) Joint scheduling decision: In the dynamic priority scheduling step, when a route adjustment instruction is generated for the transport machinery, the traffic impedance will be considered. Incorporate this into the total scheduling cost calculation. The scheduling optimization objective function is: ,in, This is the increase in the total construction period. For traffic impedance, For the potential benefits of early completion, , These are the weighting coefficients.

[0071] The final scheduling scheme satisfies the dual constraints of optimal construction progress and controllable traffic impact.

[0072] This embodiment provides an intelligent municipal engineering construction progress management method that can significantly improve the coordination efficiency and progress control capability of municipal engineering construction without increasing the investment in a large amount of fixed infrastructure.

[0073] Example 2 like Figure 2 As shown, an intelligent municipal engineering construction progress management method, based on Embodiment 1, further introduces a flexible planning function for process time windows based on historical conflict pattern learning. Embodiment 1 solved the problem of real-time avoidance when machine groups are about to conflict spatially, while this embodiment focuses on a more advanced stage—actively adjusting the start time and operation rate of each process in the daily rolling schedule of the construction plan, reducing the frequency of structural conflicts between machine groups from the source, so that the distributed negotiation mechanism only needs to handle residual occasional conflicts. Specifically, it includes the following steps: The following sub-steps are added to step 3: Regardless of whether automatic arbitration is reached during the negotiation process, the mobile terminal will package the structured record of this conflict event—including the mechanical identification and type of both parties, the station number and timestamp of the conflict, the type of conflict, the overlapping area, the work process type and priority of both parties at that time, the negotiation result (which party yields or reports to the cloud) and the actual waiting time—into a conflict log and upload it to the cloud platform through the communication link.

[0074] The generation and uploading of this log are real-time and automatic, embedded at the end of the distributed conflict negotiation step as a new output of that step.

[0075] The following sub-steps are added in step 4: A new sub-step for conflict hotspot detection is added. This sub-step runs at a preset cycle, receives accumulated conflict logs, and performs the following operations: (1) Aggregate spatiotemporal grid: Divide the construction line into equidistant spatial grid units according to the station number (e.g., 10 meters / grid), divide the time axis into equal time slots according to the operation period (e.g., 15 minutes / slot), and count the frequency of various conflicts and the cumulative waiting time in each time slot; (2) Hotspot scanning: Using spatiotemporal scanning statistical methods, high-density clustering areas with conflict frequency significantly higher than random expectations are identified, and spatiotemporal distribution maps of conflict hotspots are output; (3) Supplementary attribution: Cross-compare conflict hotspot information with process scheduling records to locate the process combinations and time periods that form structural blockages, as a supplementary dimension to the attribution conclusions in the efficiency analysis results. This conclusion not only points out the upstream root cause of downstream anomalies, but also marks the spatiotemporal hotspot areas where structural conflicts occur repeatedly and their associated processes.

[0076] The following sub-steps are added to step 5: The preceding daily rolling flexible scheduling optimization sub-step is executed when the next day's construction schedule is generated after the end of each day's construction. The specific process is as follows: (1) The planned start time offset of each work process segment (the number of minutes that can be advanced or delayed) and the work rate adjustment coefficient (allowed to fluctuate within a certain range, such as 0.8 to 1.2 times the standard rate) are used as optimization variables; (2) Ensure that the process logic sequence between processes remains unchanged, the total amount of work completed in each section remains unchanged, and the number of each type of machinery used in each work period does not exceed the total available quantity; The planned space occupancy projection of each machine in each spatiotemporal grid in the scheduling scheme to be evaluated is compared grid by grid with the spatiotemporal distribution map of conflict hotspots output in the aforementioned conflict hotspot mining sub-step. A penalty term proportional to the historical conflict frequency of the grid is applied to the planned occupancy falling into the high conflict density grid. The objective function is to minimize the total penalty of the entire scheduling. (3) Solve the above problem using a mixed integer linear programming solver to generate conflict avoidance optimization for the next day's scheduling suggestion. After confirmation by the project manager, the suggestion will take effect and serve as the planning parameter for the intention generation steps of each machine terminal the next day.

[0077] Furthermore, the process start time and planned operation rate generated by the next day's scheduling optimization directly replace the baseline planning data used by the mobile terminals of each construction machine when generating intent vectors, thereby reducing foreseeable structural conflicts at the source.

[0078] This embodiment provides an intelligent municipal engineering construction progress management method that can trace conflict avoidance back from real-time response to the planning stage, and form a closed-loop collaborative mechanism between daily rolling macro-flexible scheduling and minute-level micro-distributed negotiation, enabling more comprehensive progress management.

[0079] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics of the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent.

Claims

1. An intelligent method for managing the construction progress of municipal engineering projects, characterized in that: Includes the following steps: Step 1: The mobile terminal mounted on each construction machine collects the positioning data, visual data and machine condition data of the construction machine in real time as the construction machine moves along the linear work surface, and generates progress semantic events with spatiotemporal tags locally on the mobile terminal. Step 2: Based on the current process status and mechanical motion parameters, each mobile terminal calculates the spatial occupancy projection of the construction machinery in a future preset time window in real time, generates an intent vector, and broadcasts it to the neighboring construction machinery through the direct communication link between the mobile terminals. At the same time, the progress semantic event and intent vector are uploaded to the cloud platform. Step 3: The first mechanical terminal receives the intent vector of at least one second mechanical terminal and determines whether there is a predictive conflict by detecting spatiotemporal overlap. If a conflict exists, the terminals autonomously make avoidance decisions based on preset negotiation rules and update their respective intent vectors until the conflict is resolved. The first mechanical terminal refers to a mobile terminal that acts as the active execution end for conflict detection, and the second mechanical terminal refers to another mobile terminal that acts as the receiving and analysis end for intent vectors. Step 4: The cloud platform gathers the progress semantic events of the entire line, constructs a spatiotemporal diagram of the construction process, uses graph neural networks to learn the normal process connection pattern, identifies abnormal progress nodes and attributes them to the spatiotemporal conflict events of the machine group that caused the abnormality, and outputs the efficiency analysis results. Step 5: Based on the efficiency analysis results, the cloud platform dynamically marks the critical path process resources that affect the overall project duration and sends priority tags to the corresponding mobile terminals. The negotiation rules adjust the yielding order of each construction machine in conflict negotiation based on priority labels, thereby realizing rule-guided dynamic resource allocation.

2. The intelligent municipal engineering construction progress management method according to claim 1, characterized in that, In step 1, the positioning data is obtained by fusing RTK-GPS with ultra-wideband UWB positioning anchors deployed along the work zone to maintain centimeter-level continuous positioning in areas where GPS signals are blocked; the visual data is collected by an industrial camera facing the work mechanism, and the machine condition data is obtained through a CAN bus or an external inertial measurement unit.

3. The intelligent municipal engineering construction progress management method according to claim 1, characterized in that, The construction machinery includes pavers, road rollers, excavators, and dump trucks; the progress semantic events are generated by reasoning visual data using a lightweight convolutional neural network built into the mobile terminal, including at least: identifying the state of the material pile in front of the paver and the paving distance, identifying the texture changes associated with the number of compaction passes of the road roller, and identifying the number of cycles of the excavator loading action. The progress semantic events are associated with the stationing information to form a progress profile that is continuously distributed along the linear project.

4. The intelligent municipal engineering construction progress management method according to claim 1, characterized in that, The intent vector includes at least: machine identifier, machine type, space occupancy area described by polygons or trajectories, time window in which the space occupancy area is effective, and the initial priority of the current operation of the construction machine.

5. The intelligent municipal engineering construction progress management method according to claim 1, characterized in that, In step 3, the spatiotemporal overlap detection simultaneously determines the geometric intersection in space and the overlap in time windows; the predictive conflict classification is divided into hard conflicts of physical collision and soft conflicts of spatial encroachment that cause operation to be hindered.

6. The intelligent municipal engineering construction progress management method according to claim 1, characterized in that, The negotiation rule is a rule-cost hybrid model based on priority and local avoidance costs: The two parties in the conflict exchange priorities and the time loss costs of their respective avoidance actions, and complete the arbitration based on the logic that the lower-priority party should yield, and the party with higher cost for the same priority has the right to demand that the lower-cost party yield.

7. The intelligent municipal engineering construction progress management method according to claim 1, characterized in that, In step 4, the construction process spatiotemporal diagram uses each work segment as a node and the machine group movement path as an edge, embedding the progress status and machine group interaction features of each time period; the graph neural network is a graph attention network, which learns the spatiotemporal pattern of normal process connection, and when an abnormal progress stagnation node occurs, it outputs the upstream machine group conflict event that caused the stagnation and its impact range through attention weight backtracking.

8. The intelligent municipal engineering construction progress management method according to claim 1, characterized in that, In step 5, the cloud platform continuously tracks the progress dependencies of each process segment and the delay propagation chain predicted by the graph neural network. Any process that will further delay the total project duration is marked as a critical path process, and the mechanical resources serving the process are identified as critical resources. The priority label includes the critical resource label.

9. The intelligent municipal engineering construction progress management method according to claim 1, characterized in that, When the cloud platform identifies a systemic bottleneck that distributed negotiation cannot resolve, i.e., when two or more processes marked as critical resources are competing for channels, it generates a macro-scheduling instruction and directly sends it to the relevant mobile terminals, forcibly adjusting the construction sequence or spatial path of at least one of them, thus breaking the deadlock from a global perspective.

10. The intelligent municipal engineering construction progress management method according to claim 1, characterized in that, The cloud platform also accesses dynamic traffic situation data from the urban road network surrounding the construction area; In dynamic resource allocation, when calculating scheduling strategies related to transportation machinery, the cloud platform incorporates the real-time traffic impedance of the path into the scheduling cost calculation, generating a joint scheduling scheme that takes into account the traffic impact.

Images