Data model driven road infrastructure digital twin simulation prediction method

Abstract

The application relates to the technical field of a data model driving-based road infrastructure digital twin simulation prediction method, and particularly discloses a data model driving-based road infrastructure digital twin simulation prediction method. The method comprises the following steps: constructing a three-dimensional digital base model integrating road geometry, traffic facilities and environmental elements; dynamically slicing and returning in a lightened mode through a cloud rendering engine; realizing spatiotemporal dynamic updating of the digital twin by accessing multi-source real-time and historical data; coupling multi-agent traffic simulation and infrastructure degradation model to deduce future operation states; and generating maintenance priority and control optimization suggestions for a decision platform. Through the technical scheme, the road system is changed from static visualization to dynamic simulation prediction, and the traffic operation efficiency and the whole life cycle management level of the infrastructure are improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent transportation systems and digital twin technology, specifically relating to a data model-driven simulation and prediction method for digital twins of road infrastructure. Background Technology

[0002] With the deepening of intelligent transportation and urban digitalization, digital twin technology has become a core means to improve the management level and safety of road infrastructure. Constructing a highly realistic digital space to achieve real-time monitoring and operational pattern exploration of the physical road system is of significant strategic importance for optimizing transportation management and service efficiency. In complex transportation network environments, achieving deep mapping and virtual-real interaction between the physical world and digital space has become a key driving force for promoting the high-quality development of the transportation industry.

[0003] Data model-driven digital twin simulation and prediction of road infrastructure is a core technological direction for achieving precise traffic control and scientific maintenance decisions. This technology aims to integrate road geometry, real-time traffic flow, and various traffic event data to construct a digital twin foundation with real-time response and forward evolution capabilities. Through the digital transcription and multi-dimensional simulation of the operational status of road infrastructure, it provides traffic management departments with comprehensive support, from macro-trend prediction to micro-maintenance recommendations, thereby improving the operational efficiency of urban traffic.

[0004] Existing road digital twin technologies are mostly limited to static display and status visualization of 3D scenes, consuming huge rendering resources and exhibiting poor compatibility with complex 3D models, resulting in lag in visualization interactions at large scales. Furthermore, in-depth research into the underlying data models is lacking; existing data analysis models primarily focus on post-event recaps of traffic incidents, lacking the ability to use digital twins for pre-simulation and prediction of the overall road system status. In addition, infrastructure maintenance decisions still rely on passive algorithm analysis based on the current situation, failing to achieve pre-optimization of maintenance plans and cost control, resulting in insufficient intelligent decision support for traffic management departments and difficulty in meeting the needs of collaborative traffic governance in highly dynamic environments.

[0005] Therefore, a data model-driven digital twin simulation and prediction method for road infrastructure is desired. Summary of the Invention

[0006] The purpose of this invention is to provide a data model-driven digital twin simulation and prediction method for road infrastructure, which can solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the technical solution adopted by this invention is as follows: a data model-driven digital twin simulation and prediction method for road infrastructure, comprising the following specific steps: Step 1: Construct a digital twin base for road infrastructure. By integrating and modeling road geometric data, traffic facility layout data, and environmental element data through a unified spatial coordinate system, a three-dimensional digital base model with multi-source heterogeneous data compatibility is formed. Step 2: Deploy the cloud rendering engine, dynamically slice the 3D digital base model according to the regional granularity, and call the model slices of the corresponding region according to the real-time interaction request of the user terminal. Perform lightweight rendering tasks through the cloud GPU cluster, and send the rendering results back to the user terminal in the form of video stream. Step 3: Integrate real-time traffic flow data, historical traffic event data, and meteorological environment data to dynamically label and update the attributes of the road infrastructure in the three-dimensional digital base model, forming a digital twin with spatiotemporal evolution capabilities; Step 4: Based on the digital twin, construct a multi-agent traffic simulation model to simulate the operation behavior of vehicles, pedestrians and non-motorized vehicles under different traffic control strategies, and combine the physical degradation law of road infrastructure to deduce the overall operation status of the road system within a specific time window in the future. Step 5: Based on the simulation results, generate a road maintenance priority sequence and traffic control optimization suggestions, and output them to the traffic management decision support platform to guide the pre-maintenance work and dynamic signal timing adjustment.

[0008] Preferably, the road geometry data in step 1 includes lane topology, slope curvature parameters and intersection channelization design information; traffic facility layout data includes traffic light locations, sign coordinates and guardrail distribution information; and environmental element data includes green belt range, drainage ditch direction and surrounding building outlines. All data are spatially aligned through a unified geocoding system to ensure geometric consistency of the model at the centimeter level.

[0009] Preferably, in step 2, the region granularity is divided into highway sections, urban arterial road sections, and intersection areas according to the road function level. Each type of region corresponds to a different model detail level strategy. The intersection area retains all traffic facility details, the arterial road section retains only lane-level geometric features, and the highway section adopts simplified road surface texture to reduce the cloud rendering load. The video stream backhaul adopts an adaptive bitrate control mechanism, which dynamically adjusts the frame rate and resolution according to the network bandwidth to ensure that the user terminal interaction latency is lower than a preset threshold.

[0010] Preferably, in step 3, the real-time traffic flow data comes from geomagnetic coils, video detectors, and floating car trajectories. Historical traffic event data includes accident records, road occupancy information, and road closure notices for large-scale events. Meteorological environmental data covers rainfall intensity, visibility, and road icing probability. After all data is aligned with timestamps, it is mapped to the corresponding spatiotemporal units in the digital twin, and dynamic values ​​are assigned to the road surface friction coefficient, traffic capacity, and facility availability.

[0011] Preferably, in step 4, the multi-agent traffic simulation model adopts a hierarchical behavior rule architecture. The upper layer is a path planning module, which generates individual travel intentions based on the real-time OD matrix. The middle layer is a car-following and lane-changing module, which simulates micro-driving behavior based on a modified intelligent driver model. The lower layer is a conflict resolution module, which handles trajectory conflicts within intersections. The physical degradation law of road infrastructure is jointly characterized by a material fatigue accumulation model and a load frequency statistical model, which is used to predict the pavement damage index and the decline trend of bridge bearing capacity.

[0012] Preferably, the road maintenance priority sequence in step 5 is calculated based on the deduced facility failure risk level, traffic impact range, and repair cost. The traffic control optimization suggestions include signal cycle adjustment, green light ratio allocation scheme, and reversible lane activation timing. All output results are encapsulated in structured data format and accompanied by confidence index, for the traffic management decision support platform to compare multiple schemes and conduct manual review.

[0013] Preferably, the digital twin supports simulation at multiple time scales: short time scales of 5 minutes to 2 hours for emergency response simulations, medium time scales of 24 hours to 7 days for routine maintenance planning, and long time scales of 30 to 180 days for infrastructure lifecycle management. Different data update frequencies and model simplification strategies are adopted at different time scales to ensure efficient utilization of computing resources.

[0014] Preferably, the method further includes a mechanism for retrospectively verifying the simulation prediction results, performing deviation analysis between the actual traffic operation status and the historical projection results, extracting the dominant error factors and feeding them back to the multi-agent behavior rule base and infrastructure degradation model, thereby realizing the self-learning and self-optimization capabilities of the digital twin and continuously improving the prediction accuracy.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By constructing a data model-driven digital twin of road infrastructure, the transformation from static visualization to dynamic simulation and prediction has been realized, solving the problems of rendering lag, poor model compatibility, and lack of pre-decision support capabilities in traditional digital twin systems; cloud rendering and dynamic slicing technologies are adopted to significantly reduce terminal hardware dependence while ensuring smooth interaction in large scenes; deep integration of multi-source heterogeneous data enables the digital twin to have the spatiotemporal evolution characteristics of the real world. 2. By coupling multi-agent simulation with infrastructure degradation model, a systematic prediction of the future state of road system was realized for the first time, providing scientific, forward-looking and quantitative support for maintenance decision-making and traffic control, and improving the operational efficiency of urban transportation system and the level of full life cycle management of infrastructure. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of simulation prediction based on the coupling of multi-agent behavior rules and infrastructure physical degradation laws in this invention; Figure 3 This is a logical flowchart of the construction of a multi-source heterogeneous data fusion and digital twin base based on a unified spatial coordinate system in this invention. Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow between the cloud rendering engine execution model dynamic slice rendering and video stream back transmission in this invention; Figure 5 This is a logical flowchart of the process of generating road maintenance priorities and traffic control suggestions based on multi-timescale simulation in this invention. Detailed Implementation

[0017] Example 1: Please refer to the appendix Figure 1 To be continued Figure 5 To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to specific embodiments.

[0018] In the data model-driven digital twin simulation and prediction method for road infrastructure, step 1 specifically involves constructing a digital twin base for road infrastructure. This is achieved by fusing road geometric data, traffic facility layout data, and environmental element data using a unified spatial coordinate system to form a three-dimensional digital base model capable of handling multi-source heterogeneous data. Specifically, in constructing the digital twin base, high-precision surveying methods are first used to acquire geospatial data of the entire road. The road geometric data includes lane topology relationships, slope curvature parameters, and intersection channelization design information. When processing lane topology relationships, the system abstracts each lane as a vector line segment with directional attributes and constructs a lane-level topology network through node connections, recording the merging, diverging, and intersection relationships between lanes. Slope curvature parameters are extracted by fitting elevation point cloud data, calculating the percentage of the road's longitudinal slope and the radius of horizontal curves to ensure that the digital twin model's vertical shape is highly consistent with the actual road surface. The intersection channelization design information covers micro-geometric features such as the distribution of guide lines, the setting of left-turn waiting areas, and the corner radius of the intersection.

[0019] The traffic facility deployment data encompasses traffic light locations, sign coordinates, and guardrail distribution information. Traffic light locations include not only the geographical latitude and longitude of the light poles but also details such as the height, orientation, and lane coverage of each light group. Sign coordinates are linked to the physical dimensions, semantic content, and supporting structure type of the signs. Guardrail distribution information records the start and end points, height, and material properties of the central median guardrail and the side roadbed guardrails. Environmental element data includes the extent of green belts, the direction of drainage ditches, and the outlines of surrounding buildings. All these heterogeneous data are spatially aligned using a unified geocoding system. The system uses the National Geodetic Coordinate System as a reference, uniformly converting surveying drawings, satellite imagery, laser point clouds acquired by mobile surveying systems, and building information model data from different sources to the same projection plane. Multi-source data fusion algorithms eliminate spatial observation errors between different sensors, ensuring geometric consistency of the model at centimeter-level accuracy.

[0020] In the above method, step 2 involves deploying a cloud rendering engine to dynamically slice the 3D digital base model at the regional granularity. Based on real-time interaction requests from the user terminal, the system calls the corresponding regional model slices, executes lightweight rendering tasks via a cloud GPU cluster, and transmits the rendering results back to the user terminal as a video stream. Specifically, the regional granularity is divided into highway sections, urban arterial road sections, and intersection areas according to road function levels, with each type of area corresponding to a different model detail level strategy. In intersection areas, due to complex traffic flow and dense facilities, the system adopts the highest level of detail, retaining all traffic facility details, including the structure of traffic light heads, subtle wear textures on the ground, and fastener models of poles. In urban arterial road sections, to balance the rendering load, the system adopts a medium level of detail, retaining only lane-level geometric features, central dividers, and major directional signs. For highway sections, due to obvious linear features and high scene repetition, the system simplifies road surface textures and performs significant patching of distant buildings.

[0021] When executing rendering tasks, the cloud-based GPU cluster dynamically allocates computing resources based on the current number of concurrent users and the viewport range. When a user performs zooming, panning, or rotating operations on the terminal, the terminal encapsulates the real-time interaction request into coordinate instructions and sends them to the cloud. The cloud rendering engine quickly identifies model slices within the view frustum range based on the instructions and triggers the rendering pipeline to generate images. Video stream return employs an adaptive bitrate control mechanism, which monitors the round-trip latency and packet loss rate of the network transport layer in real time. If bandwidth constraints or increased latency are detected, the system automatically reduces the encoding resolution or frame rate to prioritize smooth interaction; when network conditions improve, it gradually restores high-definition video output. Through this adaptive adjustment, the user terminal interaction latency is ensured to be below a preset threshold of 100 milliseconds, thereby achieving seamless loading and smooth operation of the large-scene digital twin platform.

[0022] In the above method, step 3 specifically involves accessing real-time traffic flow data, historical traffic event data, and meteorological environmental data to dynamically label and update the road infrastructure status and attributes in the three-dimensional digital base model, forming a digital twin with spatiotemporal evolution capabilities. Real-time traffic flow data originates from geomagnetic coils, video detectors, and floating car trajectory data distributed along the road. Geomagnetic coils provide accurate headway and occupancy data; video detectors extract queue lengths and vehicle type ratios using target recognition algorithms; and floating car trajectories provide average speed and travel time for the entire road segment. Historical traffic event data includes accident records, road occupancy information, and road closure notices for large-scale events. The system automatically extracts the spatial station number, duration, and affected lane index of the event by parsing structured logs from traffic management departments. Meteorological environmental data covers rainfall intensity, visibility, and the probability of road icing; this data is collected in real-time by automatic weather stations deployed along the route.

[0023] After all multi-source data is aligned with timestamps, it is mapped to corresponding spatiotemporal units in the digital twin using spatiotemporal indexing technology. Specifically, the system divides the road surface into spatial grids with fixed step sizes and dynamically calculates the friction coefficient of the road surface based on rainfall intensity and temperature data. When the rainfall intensity reaches a certain level, the system adjusts the traffic capacity correction coefficient of that road segment downwards. Simultaneously, the availability of traffic facilities is also dynamically assigned. For example, if the video detects a lighting malfunction in a certain road segment, the corresponding entity attribute in the digital twin will be marked as faulty, and the lighting representation in the rendering layer will be changed in real time. In this way, the 3D digital base is transformed from a static model into a digital twin capable of reflecting the dynamic evolution of the real-world physical state.

[0024] In the above method, the core of step 4 lies in constructing a multi-agent traffic simulation model based on the digital twin. This model simulates the operational behavior of vehicles, pedestrians, and non-motorized vehicles under different traffic control strategies and, combined with the physical degradation patterns of road infrastructure, predicts the overall operational status of the road system within a specific future time window. The multi-agent traffic simulation model adopts a hierarchical behavioral rule architecture. The upper layer is the path planning module. Based on the real-time Origin-Destination matrix (i.e., the travel demand matrix from each origin to the destination), the system generates individual travel intentions for each traffic participant and dynamically searches for the optimal path based on real-time traffic conditions. The middle layer is the car-following and lane-changing module. Car-following behavior is based on a modified intelligent driver model. In this model, the vehicle's acceleration is determined by the ratio of expected speed to actual speed and the ratio of expected distance to actual distance. Different driving styles are simulated by adjusting parameters such as comfort acceleration and safe braking deceleration. Lane-changing behavior comprehensively considers the speed gain and lane-changing safety of the target lane. A lane-changing action is triggered only when the expected acceleration of the target lane is greater than that of the current lane and the following distance after the lane change meets the safety threshold. The lower layer is the conflict resolution module, which is specifically designed to handle trajectory conflicts within intersections or merging zones. It coordinates the passage order of different agents through first-come-first-served logic or priority judgment based on right-of-way.

[0025] The physical degradation of road infrastructure is characterized by a combination of a material fatigue accumulation model and a load frequency statistical model. The system uses a digital twin to statistically analyze the axle load distribution of vehicles passing through each road segment in real time, and calculates the cumulative fatigue damage borne by the pavement structural layers based on the equivalent single-axle load conversion principle. The pavement damage index prediction combines the initial pavement strength, service life, and the aforementioned cumulative damage, simulating the development trends of cracks, rutting, and potholes through attenuation equations. The bridge load-bearing capacity degradation trend focuses on the calculation of steel corrosion models and concrete carbonation depth, extrapolating the decline in structural resistance under specific traffic loads. By inputting the dynamic loads generated by multi-agent simulations into the physical degradation model in real time, the system can systematically extrapolate the future overall state of the road system.

[0026] In the above method, step 5 involves generating a road maintenance priority sequence and traffic control optimization suggestions based on the simulation results, and outputting them to the traffic management decision support platform. The generation of the road maintenance priority sequence is a multi-objective optimization decision-making process. The system performs comprehensive calculations based on the simulation results of facility failure risk level, traffic impact range, and repair cost. The failure risk level is determined by the pavement damage index or structural safety index; the traffic impact range is quantified by simulating the total traffic delay caused by facility failure or road closure in a digital twin; and the repair cost is estimated based on engineering quantity quotas. The system performs a weighted summation of the above three dimensions and outputs a maintenance task list from high to low.

[0027] Traffic management optimization suggestions include signal cycle adjustment, green light ratio allocation schemes, and the timing of reversible lane activation. The system conducts parallel simulation tests on multiple control plans in a digital twin environment. For signal cycle adjustment, the system searches for the optimal cycle length with the goal of minimizing total intersection delay; for the green light ratio, it dynamically divides the traffic flow based on the real-time flow distribution of each phase. The activation logic for reversible lanes is based on the imbalance of lane occupancy; when traffic flow in one direction approaches saturation while the opposite direction's flow is extremely low, a trigger command to activate the reversible lane is generated. All output results are encapsulated in a structured data format and include a confidence index. The confidence level is a percentage value determined by the degree of fit between the simulation model and actual observation data, as well as the completeness of the input data. The traffic management decision support platform uses this information to compare multiple schemes and conduct manual review, guiding actual maintenance operations and dynamic signal adjustments.

[0028] The digital twin supports multi-timescale simulations to meet the needs of different business scenarios. Short-term simulations, typically set from 5 minutes to 2 hours, are primarily used for emergency response simulations of sudden events. For example, when a traffic accident causes partial lane closures, the system immediately initiates a short-term simulation to predict the congestion spread trend in the affected area within the next hour. Medium-term simulations, set from 24 hours to 7 days, are used for developing routine maintenance plans. The system simulates traffic flow fluctuations over the next week to identify construction windows with minimal traffic impact. Long-term simulations, ranging from 30 to 180 days, focus on the entire lifecycle management of infrastructure, predicting the pavement performance degradation trajectory over the next six months. The system employs differentiated data update frequencies for different timescales. Short-term simulations use real-time traffic flow updates at the second level, while long-term simulations rely more on analytical calculations of macroscopic traffic demand models and physical degradation models, reducing redundant consumption of computational resources through model simplification strategies.

[0029] The method also includes a mechanism for backtracking and verifying simulation prediction results, which is a crucial closed loop for achieving system self-evolution. The system periodically compares actual traffic conditions, such as actual congestion duration and actual traffic volume, with historical projections, calculating the deviation between the two. Through deviation analysis, the system can identify the dominant factors causing prediction errors, such as improper setting of car-following behavior parameters for a certain type of vehicle, or deviations in the weather sensitivity coefficient in the road surface attenuation equation. These analysis results are fed back into the multi-agent behavior rule base and infrastructure degradation model, where machine learning algorithms automatically adjust the parameter weights within the model, enabling the digital twin to learn itself and continuously improve accuracy in subsequent predictions.

[0030] Example 2: In another specific application scenario, the method of the present invention is deployed in a complex urban transportation hub area containing large overpasses and underground tunnels. In constructing the digital twin base for road infrastructure, step 1 involves detailed modeling of the complex multi-layered spatial structure. Since underground tunnels cannot receive GPS signals, the system introduces mobile mapping technology based on the fusion of lidar point clouds and inertial navigation to acquire cross-sectional geometric data and the location of electromechanical equipment inside the tunnel. Facilities such as lighting fixtures, jet fans, and fire hydrants within the tunnel are modeled as independent digital twin entities, and their rated power, service life, and real-time operating status attributes are associated. For multi-layered overpasses, the system establishes a high-precision bridge structural model, defining in detail the spatial topological relationships of main beams, piers, supports, and expansion joints to ensure that the bridge accurately reflects the physical response of the structure under stress during simulation.

[0031] In step 2, the cloud rendering engine optimized its rendering strategy for the specific perspectives of tunnels and overpasses. In the tunnel scene, to simulate realistic visual effects and reduce computational overhead, the system adopted a simplified lighting model based on ray tracing, focusing on rendering ground reflections and the gradations of light and shadow on lamps, while using multi-level progressive texture mapping technology for the tunnel walls. In the overpass area, due to the drastic changes in viewpoint height, the cloud rendering engine employed a view distance clipping algorithm, dynamically adjusting the number of faces in the model based on the actual distance from the camera to the bridge components. When the user's viewpoint is at the bottom level of the road surface, the model of the top ramp automatically switches to a low-resolution placeholder; when the viewpoint zooms out, an asynchronous loading mechanism completes the model details, ensuring a stable rendering frame rate even in complex, multi-layered scenes.

[0032] In the data access phase of step 3, for this hub area, the system additionally accesses data from the tunnel environmental monitoring system, including carbon monoxide concentration, visibility coefficient, and wind speed and direction. This environmental data is mapped to the tunnel digital twin through specific transformation logic. For example, when the carbon monoxide concentration rises to a preset first-level warning threshold, the tunnel status attributes in the digital twin are automatically updated, causing a decrease in the expected vehicle speed in the simulation, reflecting the driver's cautious mindset in adverse environments. Simultaneously, the system accesses real-time data from the dynamic weighing system on the overpass to obtain the actual loads of passing heavy trucks. This load data is accumulated in real-time into the stress cycle count of the corresponding bridge components.

[0033] In step 4 of the simulation, for the tunnel scenario, the multi-agent traffic simulation model incorporated special vehicle behavior logic and emergency avoidance rules. The system simulated the emergency braking and avoidance paths of following vehicles when a disabled vehicle blocked the lane in a tunnel. For overpasses, the system, combining the principles of structural finite element analysis, simulated the eccentric pressure exerted on the supports by heavy-duty vehicles traveling on specific ramps within a purely text-based logical framework. By calculating the functional relationship between the number of heavy-duty vehicles passing per unit time and the decrease in bridge fatigue resistance, the probability distribution of the risk of excessive support displacement within the next three months was derived.

[0034] In the decision output stage of step 5, the generated maintenance priority sequence not only considers pavement damage but also delves into the tunnel's electromechanical system. System simulations revealed that due to the recent continuous high temperatures, the tunnel ventilation system's motors have been operating at high loads for an extended period, with a predicted probability of over 80% for thermal failure within the next 15 days. Therefore, ventilation system inspection is prioritized in the maintenance sequence. Traffic control recommendations include contingency plans for this hub. When simulations predict that mainline traffic is about to overflow onto ramps, causing interchange paralysis, the system generates recommendations including posting diversion information on electronic guidance screens 300 meters in advance and extending the red light duration on upstream entrance ramps. These proactive control measures effectively reduce the operational risks of this complex hub area.

[0035] In this embodiment, the backtracking verification mechanism utilizes a comprehensive video surveillance system within the hub for data benchmarking. The system automatically extracts traffic flow, average lane speed, and queue length from the video and performs a pixel-level spatial comparison with the results previously extrapolated using a digital twin. If the actual queue length of a certain ramp is found to be consistently greater than the extrapolated length, the system analyzes whether the diversion behavior parameters at that location are overly optimistic. By adjusting the acceptance gap parameters for ramp diversion in the multi-agent model, subsequent simulations can more realistically simulate the traffic operation characteristics of the hub.

[0036] Example 3: The method of this invention is applied to long sections of intercity highways, focusing on road protection and maintenance under severe winter weather conditions. In step 1, since the total length of the road section exceeds 100 kilometers, the digital twin base was constructed using large-scale automatic scene generation technology based on satellite remote sensing imagery, combined with oblique drone photography data from key road sections. Environmental data collection focused on slope aspect information of sections prone to snow accumulation and the physical location of snow barriers, as these geometric features directly affect snow distribution and melting rate after snowfall.

[0037] In step 2, due to the wide coverage of the long road segment, the cloud rendering engine adopted a distributed storage solution based on geographic tiles. The system divides the 100-kilometer road segment into thousands of geographic tiles, each containing independent terrain, road surface, and ancillary facility data. When a user quickly scrolls through the page on their device, the cloud-based load balancer redirects the requested tile number to the rendering node storing the corresponding tile data, achieving smooth switching between scenes across different regions.

[0038] In step 3, for the winter operating environment, the system accesses real-time road condition sensor data distributed at key points along the road section. These sensors can distinguish whether the road surface is dry, wet, covered in snow, or icy. Meteorological environmental data is refined to hourly snowfall forecasts for the next 48 hours. The system establishes a road snow depth prediction logic based on the energy balance principle, using snowfall intensity, air temperature, solar radiation, and heat release from vehicle compaction as input variables to dynamically simulate the spatiotemporal evolution of road snow thickness in a digital twin.

[0039] In the simulation phase of step 4, the multi-agent model adjusted the vehicle's dynamic parameters for icy and snowy roads. The system directly correlated the reduction in the road friction coefficient with the safe following distance parameter of the car-following model. In the verbal logical description, the vehicle's minimum braking distance was set to be inversely proportional to the road friction coefficient, meaning that on icy roads, the agent would automatically maintain a following distance three times greater than normal. Simultaneously, the physical degradation model focused on analyzing the chemical corrosion effects of de-icing agents on road materials and bridge structures. The system statistically analyzed the frequency of snow removal operations and the amount of salt applied, extrapolating the degree to which de-icing agents penetrated into the road structure layer and weakened the asphalt adhesion performance, thereby predicting the likelihood of concentrated pothole outbreaks after snowmelt in the spring.

[0040] In the output of step 5, the generated road maintenance priority sequence is transformed into winter emergency snow removal scheduling suggestions. Based on the snow depth projections, the system generates optimized routes for snowplows and control commands for the amount of de-icing agent distributed. The management optimization suggestions focus on speed limit management. Based on the simulated safe braking distance, the system calculates the maximum safe speed under current road conditions and pushes speed adjustment commands to highway variable speed limit signs, such as reducing the speed from 120 km / h to 60 km / h, while simultaneously generating overtaking restrictions for large vehicles.

[0041] Through multi-timescale simulations, the system analyzes the impact of severe winter weather on the annual road maintenance budget over a long timescale, predicting the reduction in road lifespan due to intensive snow removal in winter, and providing a quantitative basis for the infrastructure overhaul plan for the following year. The backtracking verification mechanism verifies the system's accuracy in identifying high-risk snow-covered road sections by analyzing the correlation between accident incidence and predicted risk levels. If actual accident hotspots are not in the predicted high-risk sequence, the system automatically performs source analysis to determine if crosswind characteristics or poor drainage leading to localized icing have been overlooked, and adjusts the environmental attribute parameters of the digital twin accordingly.

[0042] In the above embodiments, the core logic of data model-driven approaches lies in establishing a closed-loop process from perception and recognition to simulation calculation and decision output. The digital twin is no longer merely a static, visualized image, but has become a digital experimental field capable of perceiving the past, simulating the present, and predicting the future. Cloud rendering technology solves the problem of displaying large-scale, high-precision models on ordinary terminals, enabling complex traffic simulations and physical degradation predictions to be presented to managers in an intuitive video stream format.

[0043] When processing road infrastructure status labeling and attribute updates, the system employs a multi-layered attribute overlay mechanism. The base model not only stores geometric vertices and texture information but also defines semantic data containers. For example, the semantic attributes of pavement units include material type, design life, cumulative standard axle traffic, current smoothness index, and real-time surface moisture content. When multi-source data from step 3 flows in, the system converts sensor readings into semantic attribute update instructions using preset parsing rules. This data-driven mechanism ensures that simulation predictions are based on a real-time, accurate physical state benchmark.

[0044] In constructing a multi-agent traffic simulation model, to achieve a high degree of realism, the system incorporates textual representations of drivers' psychophysical characteristics. For example, when simulating nighttime driving behavior, the system dynamically adjusts the agents' line-of-sight parameters based on the distribution of lighting facilities in the digital twin. When the light intensity falls below a certain value, the agent's reaction time to changes in the speed of vehicles ahead is increased by a set amount. This fine-tuning of micro-parameters allows the simulation results to more accurately reflect real-world traffic patterns.

[0045] Modeling the physical degradation patterns is not limited to road surfaces but extends to ancillary facilities such as drainage systems and retaining walls. The system simulates the water-carrying capacity of drainage ditches under heavy rainfall conditions, extrapolating how poor drainage due to siltation leads to an increase in subgrade moisture content. Then, using a subgrade strength attenuation model, it calculates the decreasing trend of the road section's bearing capacity. The physical degradation of retaining walls considers seasonal variations in earth pressure and the development of cracks in the wall material, extrapolating the probability of collapse under extreme rainfall events.

[0046] When generating traffic control optimization suggestions, the system employs a confidence-based multi-plan selection logic. Dozens of micro-simulation tasks run simultaneously in the cloud, each representing a combination of control plans, such as different signal cycle ratios or different speed limit gradients. The system evaluates the output results of each task, including indicators such as the percentage improvement in road network efficiency and the reduction in the number of traffic conflict points. The final suggestion output to the user not only includes the operational steps of the optimal plan but also informs the user of the plan's robustness under different traffic demand fluctuations through textual descriptions, i.e., the confidence level.

[0047] The self-learning mechanism encompassed in this invention constructs a closed-loop technology for learning from experience through continuous deviation analysis. The system can automatically identify the impact patterns of environmental changes on traffic flow; for example, drivers' route selection preferences may shift during specific holidays. This pattern is transformed into heuristic weight adjustments in the route planning module through a feedback mechanism, improving the accuracy of subsequent simulation predictions at similar time points. This self-optimization capability based on data backtracking is crucial for ensuring the long-term high reliability of the digital twin system.

[0048] Furthermore, this invention fully considers the compatibility of multi-source heterogeneous data. The digital base established in step 1 can not only receive standard geographic information data, but also parse non-standard construction log files or manual inspection reports. The system extracts the description of defects in the inspection reports, such as "the road surface crack is about 5 mm wide and about 2 meters long," using natural language processing technology, and converts it into the geometric attributes and physical damage parameters of the corresponding coordinate points in the digital twin. This comprehensive data compatibility capability enriches the input sources for simulation prediction and enhances the system's ability to characterize complex real-world situations.

[0049] In summary, this invention provides a complete technical solution from road perception to intelligent decision-making by constructing a high-precision digital twin foundation and combining efficient cloud rendering with deep coupling of multiple intelligent agents and physical degradation models. This not only enhances the traffic management department's ability to predict and handle sudden traffic events, but also provides scientific, proactive, and quantitative decision support for the full lifecycle management of road infrastructure, which has profound significance for reducing maintenance costs and improving urban traffic efficiency.

[0050] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A data model-driven digital twin simulation and prediction method for road infrastructure, characterized in that, Includes the following steps: Construct a digital twin base for road infrastructure, and integrate road geometric structure data, traffic facility layout data and environmental element data through a unified spatial coordinate system to form a three-dimensional digital base model with multi-source heterogeneous data compatibility. Deploy a cloud rendering engine to dynamically slice the 3D digital base model at the regional granularity, and call the model slices of the corresponding region according to the real-time interaction request of the user terminal. Perform lightweight rendering tasks through the cloud graphics processor cluster, and send the rendering results back to the user terminal in the form of video stream. By accessing real-time traffic flow data, historical traffic event data, and meteorological environmental data, the status of road infrastructure in the three-dimensional digital base model is dynamically labeled and its attributes are updated, forming a digital twin with spatiotemporal evolution capabilities. Based on the digital twin, a multi-agent traffic simulation model is constructed to simulate the operational behavior of vehicles, pedestrians, and non-motorized vehicles under different traffic control strategies. Combined with the physical degradation law of road infrastructure, the overall operational status of the road system within a specific time window in the future is deduced. Based on the simulation results, a road maintenance priority sequence and traffic control optimization suggestions are generated and output to the traffic management decision support platform.

2. The method for simulation and prediction of road infrastructure digital twins based on data model-driven approaches according to claim 1, characterized in that, The process of building a digital twin foundation for road infrastructure includes: Use surveying and mapping methods to obtain geospatial data of the entire road; The road geometry data includes lane topology, slope curvature parameters, and intersection channelization design information. Each lane is abstracted into a vector line segment with directional attributes, and a lane-level topology network is constructed through node connection relationships to record the merging, diverging, and intersection relationships between lanes. The traffic facility layout data includes traffic light locations, sign coordinates, and guardrail distribution information. The traffic light locations are detailed down to the height, orientation, and lane coverage of the light units. The guardrail distribution information records the start point, end point, height, and material properties of the central median guardrail and the side roadbed guardrails. The environmental element data includes the green belt range, the direction of the drainage ditch and the outline of the surrounding buildings; All heterogeneous data are spatially aligned using a unified geocoding system. The national geodetic coordinate system is used as the benchmark. Data from different sources, including surveying drawings, satellite imagery, laser point clouds and building information model data, are uniformly converted to the same projection plane. Spatial observation errors between different sensors are eliminated through multi-source data fusion algorithms.

3. The method for simulation and prediction of road infrastructure digital twins based on data model-driven approaches according to claim 2, characterized in that, The process of deploying the cloud rendering engine includes: dividing the area granularity into highway sections, urban arterial road sections, and intersection areas according to the road function level, with each type of area corresponding to a different model detail level strategy; using the highest level of detail in intersection areas, preserving all traffic facility details, including traffic light head structures, ground wear textures, and fastener models of poles; using a medium level of detail in urban arterial road sections, preserving lane-level geometry, central dividers, and major directional signs; using simplified road surface textures in highway sections and applying patch processing to distant buildings; cloud image. The shape processor cluster dynamically allocates computing resources based on the current number of concurrent users and the viewport range. When users perform zoom, pan, or rotate operations, the real-time interactive requests are encapsulated as coordinate instructions and sent to the cloud. The cloud rendering engine identifies model slices within the view frustum range based on the instructions and triggers the rendering pipeline. The video stream return adopts an adaptive bitrate control mechanism, which monitors the round-trip latency and packet loss rate of the network transmission layer in real time. When bandwidth is limited or latency increases, the encoding resolution or frame rate is automatically reduced, and high-definition video output is restored when network conditions improve, ensuring that the user terminal interaction latency is below a preset threshold.

4. The method for simulation and prediction of road infrastructure digital twins based on data model-driven approaches according to claim 3, characterized in that, The process of forming a digital twin with spatiotemporal evolution capabilities includes: accessing real-time traffic flow data, which comes from geomagnetic coils, video detectors, and floating car trajectory data along the road. The geomagnetic coils provide headway and occupancy data, the video detectors extract the queue length and vehicle type ratio of vehicles through target recognition algorithms, and the floating car trajectories provide the average driving speed and travel time of the entire road segment. Access historical traffic event data, including accident records, road closure information for construction, and road closure notices for large-scale events. By parsing the structured logs of the traffic management department, extract the spatial station number, duration, and affected lane index of the event. Meteorological environmental data is accessed, which includes rainfall intensity, visibility, and the probability of road icing. This data is collected in real time by automatic weather stations deployed along the route. After all multi-source data are aligned with timestamps, they are mapped to the corresponding spatiotemporal units in the digital twin using spatiotemporal indexing technology. The road surface is divided into a spatial grid with a fixed step size. The friction coefficient of the road surface is dynamically calculated based on rainfall intensity and temperature data. When the rainfall intensity reaches a preset level, the traffic capacity correction coefficient of the road section is adjusted downward. The availability of transportation facilities is dynamically assigned. When a lighting facility malfunction is detected, the corresponding entity attribute in the digital twin is marked as invalid, and the lighting performance in the rendering layer is changed.

5. The method for simulation and prediction of road infrastructure digital twins based on data model-driven approaches according to claim 4, characterized in that, The process of constructing a multi-agent traffic simulation model includes: adopting a hierarchical behavioral rule architecture, with the upper layer being a path planning module that generates individual travel intentions for each traffic participant based on the real-time travel demand matrix from origin to destination, and dynamically searches for the optimal path based on real-time traffic conditions; the middle layer being a car-following and lane-changing module, where car-following behavior is based on a modified intelligent driver model, setting the vehicle's acceleration to be jointly determined by the ratio of expected speed to actual speed and the ratio of expected distance to actual distance, and simulating different driving styles by adjusting comfort acceleration and safe braking deceleration parameters; lane-changing behavior comprehensively considers the speed gain and lane-changing safety of the target lane, triggering a lane-changing action when the expected acceleration of the target lane is greater than that of the current lane and the following distance after lane-changing meets the safety threshold; and the lower layer being a conflict resolution module that handles trajectory conflicts within intersections or merging zones, coordinating the passage order of different agents through first-come-first-served logic or priority judgment based on right-of-way.

6. The method for simulation and prediction of road infrastructure digital twins based on data model-driven approaches according to claim 5, characterized in that, The characterization process of the physical degradation law includes: using a material fatigue accumulation model and a load frequency statistical model for joint characterization, and real-time statistical analysis of the axle load distribution of vehicles passing through each road segment in the digital twin; calculating the cumulative fatigue damage borne by the pavement structure layer according to the equivalent single axle load conversion principle; predicting the pavement damage index by combining the initial strength, service life and cumulative damage of the pavement, and simulating the development trend of cracks, ruts and potholes through attenuation equations; extrapolating the decline process of structural resistance under specific traffic loads by calculating the steel corrosion model and concrete carbonation depth to predict the bridge bearing capacity attenuation trend; and inputting the dynamic load generated by multi-agent simulation into the physical degradation model in real time to extrapolate the future overall state of the road system.

7. The method for simulation and prediction of road infrastructure digital twins based on data model-driven approaches according to claim 6, characterized in that, The process of generating a road maintenance priority sequence includes: based on a multi-objective optimization decision-making process, a comprehensive calculation is performed based on the derived facility failure risk level, traffic impact range, and repair cost; the failure risk level is determined by the pavement damage index or structural safety index; the traffic impact range is quantified by simulating the total traffic delay caused by facility failure or road closure in a digital twin; the repair cost is estimated based on engineering quantity quotas; and a weighted sum is performed on the three dimensions of facility failure risk level, traffic impact range, and repair cost to output a maintenance task list from high to low.

8. The method for simulation and prediction of road infrastructure digital twins based on data model-driven approaches according to claim 7, characterized in that, The process of generating traffic control optimization suggestions includes: the content of the optimization suggestions includes signal cycle adjustment, green ratio allocation scheme, and the timing of variable lane activation; parallel simulation tests of multiple control plans are conducted in a digital twin environment; for signal cycle adjustment, the optimal cycle length is searched with the goal of minimizing the total intersection delay; for green ratio, dynamic segmentation is performed based on the real-time flow distribution of each phase; for the activation of variable lanes, based on the imbalance of lane occupancy, a trigger command for activating variable lanes is generated when the traffic flow in one direction is close to saturation and the flow in the opposite direction is lower than a preset ratio; all output results are encapsulated in a structured data format and accompanied by a confidence index, which is a percentage value determined by the degree of fit between the simulation model and the actual observation data and the completeness of the input data.

9. The method for simulation and prediction of road infrastructure digital twins based on data model-driven approaches according to claim 8, characterized in that, The method also includes steps to support multi-timescale simulation: Short timescales are used for emergency response simulations of sudden events. They are activated when a traffic accident causes the closure of some lanes and predict the congestion spread trend in the affected area within a preset first time period. Medium time scales are used for routine maintenance planning, extrapolating traffic flow fluctuations within a pre-set second time period, and identifying construction windows with minimal traffic impact; The long-term focus is on the full life cycle management of infrastructure, predicting the pavement performance degradation trajectory within a pre-defined third time period in the future; Different data update frequencies and model simplification strategies are adopted for different time scales. High-frequency real-time traffic flow updates are used for short time scales, while long time scales rely on analytical calculations of macro traffic demand models and physical degradation models. The consumption of computing resources is reduced through model simplification strategies.

10. The method for simulation and prediction of road infrastructure digital twins based on data model-driven approaches according to claim 1, characterized in that, The method also includes a step of retrospectively verifying the simulation prediction results: periodically comparing the actual traffic operation status with the historical projection results and calculating the deviation between the two; By identifying the dominant factors causing prediction errors through deviation analysis, the dominant factors include deviations in the setting of car-following behavior parameters of vehicle models or deviations in the meteorological sensitivity coefficient in the road surface attenuation equation; the analysis results are fed back to the multi-agent behavior rule base and infrastructure degradation model, and the parameter weights inside the model are automatically adjusted through machine learning algorithms to achieve self-learning and self-optimization of the digital twin.

Images